informed-decision/informed_decision/memo.py

305 lines
10 KiB
Python
Raw Permalink Normal View History

Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
"""The Decision Memo object.
One question per memo. A memo carries the question, the requested act, the
binding level, the brief, a hashed packet, highlights pointing into it, and the
binding slice committing *which scope this act is being entered into*.
`binding.target` is the **act-scope**. It is never derived from the token's
`tenant` claim, which is a membership fact about the principal
(`GH-DEC-2026-013` §5, PR-08).
"""
from __future__ import annotations
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
import re
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
from dataclasses import dataclass, field, replace
from enum import Enum
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
#: approval-engine's digest format. We validate the shape and NEVER compute one.
APPROVAL_DIGEST_RE = re.compile(r"^sha256:[0-9a-f]{64}$")
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
class BindingLevel(str, Enum):
ACKNOWLEDGMENT = "acknowledgment"
ORGANIZATIONAL = "organizational"
AES = "aes"
QES = "qes"
class StepKind(str, Enum):
# Weak — acknowledgment only. `accept` is illegal here on purpose.
INFORM = "inform"
COMMENT = "comment"
REVIEW = "review"
ACKNOWLEDGE = "acknowledge"
# Co-sign
MITZEICHNUNG = "mitzeichnung"
APPROVE = "approve"
# Bind
SCHLUSSZEICHNUNG = "schlusszeichnung"
SIGN = "sign"
SEAL = "seal"
@property
def is_weak(self) -> bool:
return self in (StepKind.INFORM, StepKind.COMMENT, StepKind.REVIEW, StepKind.ACKNOWLEDGE)
@property
def is_bind(self) -> bool:
return self in (StepKind.SCHLUSSZEICHNUNG, StepKind.SIGN, StepKind.SEAL)
@dataclass(frozen=True)
class PacketItem:
"""A hashed document in the packet.
Field names follow ``canonicalize.normalize_packet``, which is the governed
contract the published vectors depend on ``item_id``, not ``id``. ``label``
is UI-only and is deliberately outside the hash: renaming a document does
not change what was shown.
"""
item_id: str
label: str
hash: str
@dataclass(frozen=True)
class Highlight:
"""A pointer into the packet.
Highlights are an aid to attention, never a filter. Acknowledging them does
not narrow what is bound the whole instrument is bound (PR-12/W-highlights).
"""
id: str
item_id: str
note: str
required_ack: bool = False
severity: str = "informational"
locator: dict = field(default_factory=dict)
@dataclass(frozen=True)
class Identifier:
scheme: str
value: str
@dataclass(frozen=True)
class Principal:
"""A party in the binding slice.
A dataclass rather than a dict because the canonicalizer requires
``display_name``, ``id`` and ``kind``, and a missing one is a runtime
KeyError deep inside hashing rather than a visible defect at construction.
"""
id: str
kind: str
display_name: str
role: str | None = None
identifiers: tuple[Identifier, ...] = ()
def as_document(self) -> dict:
out: dict = {"display_name": self.display_name, "id": self.id, "kind": self.kind}
if self.role:
out["role"] = self.role
if self.identifiers:
out["identifiers"] = [
{"scheme": i.scheme, "value": i.value} for i in self.identifiers
]
return out
@dataclass(frozen=True)
class Scope:
"""The act-scope. What the person is entering, not who they are."""
kind: str
id: str
label: str
environment: str | None = None
requires_new_bind: bool = False
@dataclass(frozen=True)
class BindingSlice:
"""Committed and signed. Inside ``view_hash``."""
principal: Principal
target: Scope
terms: str | None = None
justification: str | None = None
@dataclass(frozen=True)
class Hat:
"""A role. Not a scope — `hats are not scopes` (INTENT invariant)."""
id: str
label: str
elevates: bool = False
def as_document(self) -> dict:
return {"id": self.id, "label": self.label, "elevates": self.elevates}
@dataclass(frozen=True)
class Awareness:
"""Shown on the same surface, hashed separately, never signed.
Typed for the same reason ``Principal`` is: the canonicalizer requires a
shape, and a malformed one should fail at construction rather than deep
inside hashing.
Nothing here enters ``view_hash``. Defaulting a hat to last-used is required
for situational awareness and forbidden from silently entering the signed
payload (INTENT principle 11).
"""
proposed_hat: Hat | None = None
proposed_hat_source: str | None = None
available_hats: tuple[Hat, ...] = ()
situation_note: str | None = None
def as_document(self) -> dict:
out: dict = {}
if self.proposed_hat is not None:
out["proposed_hat"] = self.proposed_hat.as_document()
if self.proposed_hat_source is not None:
out["proposed_hat_source"] = self.proposed_hat_source
if self.available_hats:
out["available_hats"] = [h.as_document() for h in self.available_hats]
if self.situation_note is not None:
out["situation_note"] = self.situation_note
return out
@dataclass(frozen=True)
class Memo:
id: str
version: int
question: str
requested_act: str
binding_level: BindingLevel
brief: str
binding: BindingSlice
step_kind: StepKind
packet: tuple[PacketItem, ...] = ()
highlights: tuple[Highlight, ...] = ()
locale: str = "en"
ui_release: str = "informed-decision@0.2.0"
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
#: Co-reference to the act this memo presents.
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
approval_id: str | None = None
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
#: approval-engine's binding.digest over the five act fields, CARRIED here
#: under GH-DEC-2026-015 (activated 2026-09-10 once approval-engine stated
#: the presentation exclusion as normative and tested).
#:
#: It is referenced, never recomputed: this repository must not restate that
#: digest from its own vocabulary. When present, the act-scope stops being
#: independently canonicalized here, so the act has exactly ONE
#: canonicalization — computed by the layer that owns it.
approval_binding_digest: str | None = None
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
sealed: bool = False
def __post_init__(self) -> None:
if not self.question:
raise ValueError("a memo without a question does not render")
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
if self.approval_binding_digest is not None:
if not APPROVAL_DIGEST_RE.match(self.approval_binding_digest):
raise ValueError(
"approval_binding_digest must be approval-engine's "
"sha256:<64 hex> form; it is carried, never computed here"
)
if self.approval_id is None:
raise ValueError(
"a carried binding digest needs the approval it belongs to"
)
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
packet_ids = {p.item_id for p in self.packet}
for h in self.highlights:
if h.item_id not in packet_ids:
raise ValueError(f"highlight {h.id} points outside the packet")
@property
def required_ack_ids(self) -> frozenset[str]:
return frozenset(h.id for h in self.highlights if h.required_ack)
def next_version(self, **changes) -> "Memo":
"""A change creates version n+1. Outstanding presentations die with it."""
return replace(self, version=self.version + 1, **changes)
def binding_document(self) -> dict:
"""The document ``view_hash`` is computed over.
Note what is here and what is not: the packet and highlights are here
because they were shown; the token's ``tenant`` claim is not, because it
is not the act-scope.
"""
return {
"memo_id": self.id,
"memo_version": self.version,
"question": self.question,
"requested_act": self.requested_act,
"binding_level": self.binding_level.value,
"brief": self.brief,
"locale": self.locale,
"ui_release": self.ui_release,
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
**(
{"approval_binding_digest": self.approval_binding_digest}
if self.approval_binding_digest is not None
else {}
),
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
"packet": [{"item_id": p.item_id, "hash": p.hash} for p in self.packet],
"highlights": [
{
"id": h.id,
"item_id": h.item_id,
"required_ack": h.required_ack,
"severity": h.severity,
"locator": h.locator,
}
for h in self.highlights
],
"binding": self._binding_document(),
}
def _binding_document(self) -> dict:
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
"""The binding slice, minus whatever the carried digest already covers.
Where ``approval_binding_digest`` is present, the **act-scope** is
omitted: `target` is act material and is covered by that digest, so
canonicalizing it again here would be the partial recomputation in a
second vocabulary that GH-DEC-2026-015 exists to remove.
``principal`` is **kept**. approval-engine's `principal` is the party
*on whose behalf* the approval was issued; ours is the person being
bound the approver. Different roles, so dropping ours would remove
*who was shown this* from `view_hash` and gut the promise this
repository exists to make. Approval Engine confirmed these distinct
roles in docs/approval-claim.md (a0a6029); this field remains.
Activate nesting per GH-DEC-2026-015: view_hash carries binding.digest approval-engine met the condition. Verified here rather than taken on report: their docs/approval-claim.md carries "Presentation exclusion — GH-DEC-2026-015 §4" in normative language, and I ran tests/test_claim_contract.py::test_presentation_changes_cannot_change_the_approved_act myself — 1 passed. That test pins the digest input set from BOTH sides, and the narrowing half is what makes it real: without it a digest over four fields, or over a constant, would pass the widening half perfectly. view_hash now carries binding.digest and the act-scope is no longer independently canonicalized here, so the act has exactly one canonicalization computed by the layer that owns it. approval_binding_digest is validated for shape and refused without its approval id — it is carried, never computed. The three published vectors are unchanged: they do not carry the new key, so pick omits it. Asserted, not assumed. The cycle condition did not disappear, its protection moved — from refusing nesting to approval-engine's normative exclusion. layer.yaml carries it as cycle_condition with a test, so a future widening meets a rule rather than silence. One thing not assumed. Both gate-house and approval-engine said our binding slice canonicalizes principal and target, two of their five fields. target plainly is act material and is now dropped. But their principal is the party ON WHOSE BEHALF the approval was issued, while ours is the person being BOUND — the approver. Different roles, and dropping ours would remove who was shown this from view_hash and gut the promise. Kept it, declared principal_role_overlap open in layer.yaml, tested that changing the approver still moves view_hash, and raised it rather than silently resolving it either way. L0/L2 are unaffected: with no approval there is no digest to defer to, and test_act_scope_still_binds_when_there_is_no_carried_digest pins that. 100 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:58:22 +02:00
"""
out: dict = {"principal": self.binding.principal.as_document()}
if self.approval_binding_digest is None:
target: dict = {
"kind": self.binding.target.kind,
"id": self.binding.target.id,
"label": self.binding.target.label,
"requires_new_bind": self.binding.target.requires_new_bind,
}
if self.binding.target.environment is not None:
target["environment"] = self.binding.target.environment
out["target"] = target
Build the T08 domain core with the engine behind a seam approval-engine APPROVAL-WP-0002-T01 is still progress and its namespace has no pods, so the live end-to-end proof cannot run. Built everything that does not depend on it, with the engine behind a Protocol plus a fake carrying its real refusal semantics, so its arrival is a wiring change rather than a build. - memo.py: the Decision Memo, versions, binding document. Principal, Scope, Awareness and Hat are dataclasses rather than dicts because the canonicalizer requires a shape and a missing key should fail at construction rather than deep inside hashing — which is exactly how it failed twice while building this. Field names follow the governed canonicalizer (item_id, severity, locator): the published vectors are the contract, so the object was aligned to them rather than the reverse. - presentation.py: the sole writer of view_hash. One writer, one canonicalizer, one place to audit. Acknowledgment is an explicit method call and nothing infers it from scroll, dwell or focus. - disposition.py: verbs and guards G_NOAGENT, G_STEP, G_PRES, G_ACK, G_REASONS, G_SEALED. accept is ABSENT from weak steps rather than present-and-disabled, because a greyed-out accept still teaches the wrong model. Only accept reaches the engine; a memo return is not represented there at all. - provenance.py: claim routes per A-16. assert_human_control_dischargeable refuses a registration-supplied human, so PR-11's limitation fires at the point of use instead of sitting in a document. - evidence.py: local outbox, commitment-only records carrying the GH-DEC-2026-014 §4 existence assertion, per-class reconciliation counts, and a custody-locator guard that rejects credentialed URLs (PR-12). - approval_client.py: 409 duplicate_approver is success, 409 conflict terminal, 503 fail-closed, approval:consume refused before a token is requested. 87 tests pass, including every negative case in the Use Case Catalog and that a fail-closed outcome is recorded as a stance application with no verb field — never as a decline, because the human did not make one. T08 stays progress: the live proof is the remainder. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V3W1dQG7GFFM9d94jFx7iR Assistant: claude-code Assistant-Model: opus Assistant-Process: 1565372@bnt-lap001 Assistant-Session: 16bb2f25-b34c-49ef-8e94-5fec3567a568
2026-09-10 20:38:20 +02:00
if self.binding.terms is not None:
out["terms"] = self.binding.terms
if self.binding.justification is not None:
out["justification"] = self.binding.justification
return out
def awareness_document(self, awareness: "Awareness | None" = None) -> dict:
"""Shown on the same surface, hashed separately, never signed."""
doc = {"memo_id": self.id, "memo_version": self.version, "locale": self.locale}
if awareness is not None:
doc.update(awareness.as_document())
return doc