Require canonical proposal frontier organs

This commit is contained in:
Meta Vens 2026-08-07 18:03:56 +01:00
commit 4338b0331a
8 changed files with 99 additions and 277 deletions

View file

@ -411,7 +411,7 @@ pub fn append_modeling_repo_model_shape_context(
item.source_scope.join(" | "),
));
}
context.push_str("New nodes must reference one exact existing domain id. Every unresolved frontier item must target at least one exact existing claim id from existingClaims, or a claim id created by the same patch. Every frontier source_scope must be non-empty and contain safe relative paths in strict lexicographic ascending order with no duplicates; proposal scope hints are evidence to normalize into that canonical order, not an instruction to preserve unsafe or unsorted serialization. A proposal Evolution upsert must use status active: Mind acceptance is the admission decision, and proposed state is not routeable by Self. Use upsert_frontier only for a genuinely new frontier id not listed in existingFrontier. Use revise_frontier when changing an id listed in existingFrontier; an existing frontier id cannot be upserted again. RepoArchitecture and RepoDataflow nodes/edges may use only observed, proposed, accepted, stale, or retired lifecycle. Prefer accepted for a source-grounded current claim. Do not invent a domain because RepoModel patches have no domain mutation operation.\n");
context.push_str("New nodes must reference one exact existing domain id. Every unresolved frontier item must target at least one exact existing claim id from existingClaims, or a claim id created by the same patch. Every frontier source_scope must be non-empty and contain safe relative paths in strict lexicographic ascending order with no duplicates; proposal scope hints are evidence to normalize into that canonical order, not an instruction to preserve unsafe or unsorted serialization. A proposal Evolution upsert must use status active: Mind acceptance is the admission decision, and proposed state is not routeable by Self. Its recommended_next_organ must be exactly one routeable canonical value with exact case: Hands, Eyes, or Imagination; lowercase labels are dead documents and Mind will refuse them. Use upsert_frontier only for a genuinely new frontier id not listed in existingFrontier. Use revise_frontier when changing an id listed in existingFrontier; an existing frontier id cannot be upserted again. RepoArchitecture and RepoDataflow nodes/edges may use only observed, proposed, accepted, stale, or retired lifecycle. Prefer accepted for a source-grounded current claim. Do not invent a domain because RepoModel patches have no domain mutation operation.\n");
context.push_str("</canonical_repo_model_shape>");
Ok(context)
}
@ -1380,6 +1380,7 @@ mod tests {
);
assert!(shape.contains("strict lexicographic ascending order with no duplicates"));
assert!(shape.contains("proposal Evolution upsert must use status active"));
assert!(shape.contains("recommended_next_organ must be exactly one routeable canonical value"));
assert!(shape.contains("Prefer accepted"));
fs::remove_dir_all(&temp)?;
Ok(())

View file

@ -101,7 +101,7 @@ If you emit an `anchor`, it must include its own stable `id`, `kind`, and `targe
The launch context includes `<canonical_repo_model_shape>`. Every new node must use one exact `existingDomains` id from that packet, and repo architecture/dataflow nodes and edges must use one of the listed legal lifecycles. Never derive a domain id from the workspace path; patches cannot create domains.
When `<repo_frontier_modeling_request>` is present, echo its exact `modelingRequestId` as `repoFrontierModelingRequestId`, use `purpose.kind=incorporate_frontier_verdict` with its exact `frontierRouteId` and `soulVerdictReceiptId`, and revise only its routed frontier item to its allowed disposition. The typed request is authority; nearby telemetry, timestamps, and prose are not substitutes. When that request is absent, use `purpose.kind=evolution` and do not invent verdict-incorporation authority.
An ordinary `purpose.kind=evolution` patch without `proposalModelingContext` may mutate nodes and edges only. It must not upsert, revise, retire, or adopt frontier items. Describe an observed frontier wound in `summary`, `evidenceGaps`, or `nextSafeMove` when no typed frontier authority exists.
When the typed launch document contains `proposalModelingContext`, echo its exact `requestId` as `proposalModelingRequestId` and use `purpose.kind=evolution`. This is the one evolution frontier exception: emit exactly one `upsert_frontier` operation grounded in the proposal, target at least one exact claim id from `<canonical_repo_model_shape>.existingClaims` (or a claim created by the same patch), cite the exact `proposalId` in both top-level `evidenceIds` and the frontier item's `evidence_refs`, set the unresolved frontier `status` to `active`, and normalize the proposal's relevant `scopeHints` into a non-empty `source_scope` of safe relative paths in strict lexicographic ascending order with no duplicates. Choose `recommended_next_organ` from the actual unresolved work; a user proposal is not automatically Hands-ready. You may also update source-grounded anatomy, but must not emit any other frontier operation. Read only that immutable typed projection against its exact canonical `modelRevision` and `modelHash`. The proposal is evidence, not campaign truth: Mind still owns admission, and Hands still owns execution. Do not substitute adjacent proposals, infer service intake, parse proposal authority from prose, or claim admission or execution authority.
When the typed launch document contains `proposalModelingContext`, echo its exact `requestId` as `proposalModelingRequestId` and use `purpose.kind=evolution`. This is the one evolution frontier exception: emit exactly one `upsert_frontier` operation grounded in the proposal, target at least one exact claim id from `<canonical_repo_model_shape>.existingClaims` (or a claim created by the same patch), cite the exact `proposalId` in both top-level `evidenceIds` and the frontier item's `evidence_refs`, set the unresolved frontier `status` to `active`, and normalize the proposal's relevant `scopeHints` into a non-empty `source_scope` of safe relative paths in strict lexicographic ascending order with no duplicates. Choose `recommended_next_organ` from the actual unresolved work using exactly one canonical routeable value with exact case: `Hands`, `Eyes`, or `Imagination`; a user proposal is not automatically Hands-ready. You may also update source-grounded anatomy, but must not emit any other frontier operation. Read only that immutable typed projection against its exact canonical `modelRevision` and `modelHash`. The proposal is evidence, not campaign truth: Mind still owns admission, and Hands still owns execution. Do not substitute adjacent proposals, infer service intake, parse proposal authority from prose, or claim admission or execution authority.
Generic `statePatch` is optional and has been stripped of repository-anatomy authority. If present, it may contain only coherent `observations` and `evidence`. Never put graphs, graph frontier, graph checkpoint, scratch, investigation checkpoint, planning, objectives, jobs, or other anatomy in `statePatch`; those old paths are demoted and will be rejected.
Do not edit files, do not implement product code, do not verify as if you were the reviewer, do not change jobs/subgoals/objective/invariants/churn/mode, and do not promote or accept your own output.
'''

View file

@ -596,6 +596,7 @@ pub struct RepoFrontierPlanningCandidateEligibility {
pub frontier_item_id: String,
pub eligible: bool,
pub status_valid: bool,
pub recommended_next_organ_valid: bool,
pub source_scope_valid: bool,
pub challenged_target_claim_ids: Vec<String>,
pub unresolved_dependency_item_ids: Vec<String>,

View file

@ -3142,6 +3142,14 @@ pub fn commit_repo_model_admission(
"proposal Evolution frontier must be active so accepted unresolved work is immediately routeable"
));
}
if !matches!(
upserts[0].recommended_next_organ.as_str(),
"Hands" | "Eyes" | "Imagination"
) {
return Err(anyhow!(
"proposal Evolution frontier recommended_next_organ must be one exact routeable organ: Hands, Eyes, or Imagination"
));
}
if proposal.source_kind == crate::RepoFrontierProposalSourceKind::Imagination
&& (upserts[0].recommended_next_organ != "Imagination"
|| upserts[0].adopted_plan.is_some())
@ -6153,10 +6161,13 @@ pub fn runtime_repo_frontier_planning_eligibility(
matches!(
item.status,
crate::RepoFrontierStatus::Active | crate::RepoFrontierStatus::Proposed
) && item.recommended_next_organ == "Imagination"
) && item
.recommended_next_organ
.eq_ignore_ascii_case("Imagination")
})
.map(|item| {
let status_valid = item.status == crate::RepoFrontierStatus::Active;
let recommended_next_organ_valid = item.recommended_next_organ == "Imagination";
let source_scope_valid =
!item.source_scope.is_empty() && safe_sorted_unique_paths(&item.source_scope);
let mut challenged_target_claim_ids = challenges
@ -6184,10 +6195,12 @@ pub fn runtime_repo_frontier_planning_eligibility(
frontier_item_id: item.id.clone(),
eligible: admission_count == 1
&& status_valid
&& recommended_next_organ_valid
&& source_scope_valid
&& challenged_target_claim_ids.is_empty()
&& unresolved_dependency_item_ids.is_empty(),
status_valid,
recommended_next_organ_valid,
source_scope_valid,
challenged_target_claim_ids,
unresolved_dependency_item_ids,
@ -10611,6 +10624,7 @@ pub(crate) mod tests {
("unrelated-job", 7),
("unsorted-source-scope", 8),
("proposed-status", 9),
("noncanonical-next-organ", 10),
] {
let (store, mut result, mut review) = proposal_admission_fixture(root.path(), suffix)?;
match mutate {
@ -10665,6 +10679,19 @@ pub(crate) mod tests {
result.repo_model_patch_msgpack = Some(bytes.clone());
review.patch_sha256 = format!("{:x}", Sha256::digest(bytes));
}
10 => {
let mut patch: crate::RepoModelPatch =
rmp_serde::from_slice(result.repo_model_patch_msgpack.as_deref().unwrap())?;
let crate::RepoModelPatchOperation::UpsertFrontier { item } =
&mut patch.operations[0]
else {
unreachable!()
};
item.recommended_next_organ = "imagination".into();
let bytes = rmp_serde::to_vec_named(&patch)?;
result.repo_model_patch_msgpack = Some(bytes.clone());
review.patch_sha256 = format!("{:x}", Sha256::digest(bytes));
}
_ => unreachable!(),
}
if matches!(mutate, 6 | 7) {
@ -11124,6 +11151,7 @@ pub(crate) mod tests {
assert_eq!(eligibility.candidates.len(), 1);
assert!(!eligibility.candidates[0].eligible);
assert!(eligibility.candidates[0].status_valid);
assert!(eligibility.candidates[0].recommended_next_organ_valid);
assert!(eligibility.candidates[0].source_scope_valid);
assert_eq!(
eligibility.candidates[0].challenged_target_claim_ids,

View file

@ -17,6 +17,15 @@ eligibility projection includes unresolved proposed Imagination frontier as a
diagnostic candidate with `statusValid=false`, but only active frontier can be
eligible. This projection observes the wound; admission prevents it.
Proposal Evolution also requires exact canonical routeable organ identity:
`Hands`, `Eyes`, or `Imagination`. The field remains a string in the portable
RepoModel, but Mind admission owns the executable subset and refuses lowercase,
unknown, or merely display-shaped labels. Planning eligibility includes a
case-mismatched Imagination-shaped legacy document only to project
`recommendedNextOrganValid=false`; it never case-folds that document into
authority. v25 proved why: lowercase `imagination` was admitted but invisible to
Self's exact selector.
Owner: Mind admission owns whether proposal frontier state may enter the model.
Self only selects admitted routeable state. `source_scope` is no longer assumed
routeable merely because its individual paths are safe. Invalid ordering is

View file

@ -1,272 +1,55 @@
# Fresh Workspace Handoff
## Current authority — 2026-08-07 active-status cut
## Current authority — 2026-08-07 canonical-organ cut
Exact `717c649e403d53b12d53b4fbfabb0ecb8d49ee2d` is published as
`sha256-cec12433e99526ab8ed4795812b0bc3d91ef89d80cff324357055273733db453`
The active branch is `codex/epiphany-shakedown-live`. Exact pushed commit
`872d8ad6188fdfe8afcf5d4e9dd65a791fa444f2` is authenticated and published as
`sha256-b97f457cfde1eea2b45464b3ca12e7cdba4cb13d4c1b261b20267e960e1a326e`
with witness
`sha256-819d0141dcd20b767d0005243f884845ad1f21e95d2df188baf859bc1fbd6ae5`.
Fresh v24 was cold-started from its packaged repository-body tool at exact Git
head 717c649e; only canonical `agents.cc` crossed from v23. Its local Verse was
freshly seeded and the exact release published into it.
`sha256-40d6bda4d82a2da67614722a52810078abc37d2d5bfe60330b479fa611f726ab`.
The exact package contains 21 binaries and exposes no private state.
Packaged Self derived `launchModeling` without a proposal request override for
`repo-frontier-proposal-modeling-8b1ab2b553b7d4e37756d4cc9169a2febf68a46b5b1d8c7e15e8dd6eaf467664`.
Modeling worker `c6997b20-2978-4bac-9b7e-dccdd59ebcd7` emitted canonical sorted
scope and Mind accepted it at thread revision 3. The frontier status was
`proposed`, however, so Self returned `awaitFrontierProposal`: planning selection
only owns active unresolved frontier. Proposal admission had again accepted
state its downstream selector cannot consume.
Fresh v25 lives at
`F:\Projects\.epiphany-runtime\shakedown\live-20260807-v25`. It was cold-started
from the packaged repository-body tool at Git head `872d8ad6`; only canonical
`agents.cc` crossed from v24. Its repository Body is generation 1, its local
Verse was seeded fresh, and the exact release was published into that Verse.
The local repair requires `status=active` for every proposal Evolution frontier
upsert, tells Modeling that Mind acceptance is already the admission decision,
and projects `statusValid` alongside scope, challenge, and dependency validity.
Focused negative admission, eligibility, launch-context, and MVP status tests
pass. v24 is insert-only contaminated evidence; do not mutate or replay it.
Commit/package/publish the repair, cold-start the next generation, and repeat the
native Modeling -> Imagination -> dedicated Mind proof without overrides.
Proposal `shakedown-v25-continuity-imagination-r1` selected request
`repo-frontier-proposal-modeling-e5e04c43b3d255baa8ba2f3778e2865b8aec275d8249dca0dfa3ace034b6d80c`.
Packaged Self derived `launchModeling` without overrides. Modeling job
`5af5100d-e097-4928-9bd3-2ba60a2fbbb3` completed, and Mind accepted its active
frontier with canonical sorted unique source scope at thread revision 3.
## Current authority — 2026-08-07 routeability cut
The worker emitted `recommended_next_organ=imagination` in lowercase. Proposal
admission accepted the arbitrary non-empty string, while Self's planning
selector owns the exact canonical identity `Imagination`. Typed eligibility
therefore reported admission count one but zero candidates, and Self returned
`awaitFrontierProposal`. v25 is insert-only evidence; do not mutate or replay it.
Exact `58cf8984975c9ae524690ff2032eaa76abf5d0d4` is published as
`sha256-6c2524f1a21fce91f658ee92eb53bbe553719295fa99d7bbe5571651ce5c3c72`
with witness
`sha256-3a674c41e7560239b4437090a950def43331649995b3d51704e1687370e340e9`.
On live v23, packaged Self launched and Mind accepted fresh Modeling results for
the release-auth and Continuity proposals. They created
`frontier-imagination-release-auth-nonbypass-trace-r1` and
`frontier-imagination-continuity-crash-restart-proof-r1`; v23 reached revision
23. Self did not select either for Imagination.
The typed eligibility projection proved the exact blocker: both source scopes
are safe but not in strict lexicographic order. Planning selection already
requires non-empty safe sorted-unique paths. Proposal admission did not, so Mind
accepted state that Self can never route. The local repair adds the same
invariant to proposal admission, publishes per-frontier eligibility in MVP
status, and tells Modeling to normalize proposal scope hints into canonical
sorted-unique scope. The focused negative admission test proves rejection is
byte-preserving; the launch-context test proves the worker sees the contract.
Do not weaken the selector or add retry state. Proposal Evolution is insert-only,
so v23 has no honest typed repair for its two invalid admitted frontiers. Seal it
as evidence after the repair lands. Next: commit and push this pass, package and
publish its exact commit, bootstrap v24, then prove native
Modeling -> Imagination -> dedicated Mind without routing overrides.
## Present condition
Epiphany remains a supervised engineering alpha running cognition and release
builds on Starfire. Yggdrasil remains the small canonical public crossing body;
its current 1.9 GiB RAM does not make it a release forge.
The active pushed branch head is `2dd2534197b38df764b790bd9377374420e01d97`.
The worktree was clean before this state refresh.
Fresh v23 is the current live shakedown Body:
`F:\Projects\.epiphany-runtime\shakedown\live-20260807-v23\runtime.cc`
v23 was bootstrapped from canonical agent state plus a fresh observation of the
current repository Body. It accepted proposal
`shakedown-v23-verdict-identity-preservation-r1`, admitted RepoModel revision 1,
and granted a one-file Hands route for
`epiphany-core/src/coordinator_launch_context.rs`.
Commit `6730e395` now loads the exact current routed frontier item and renders
its identity-bearing fields into verdict-incorporation Modeling context:
`migration_body`, `question`, `target_claim_ids`, `source_scope`,
`dependency_item_ids`, `created_at`, `recommended_next_organ`, `retired_at`,
and `superseded_by`. The worker is told to preserve these exactly while only
status, evidence, gap, and update time remain verdict-owned.
The authenticated `6730e395` release is published at:
`F:\Projects\Epiphany\.epiphany-run\releases\6730e395a32c81b236cfe665f5abe70e4e89ff05\sha256-c75e33e0cf609e0dca5ba7e9a67a68bab0a71d9ff063bc52f41e92d0c6fa59d4`
Soul accepted that consequence with verdict `needs-review`: the implementation
materially closes identity preservation, but the evidence packet did not name
the existing admission-path proof. A second one-file Hands gate therefore
produced commit `2dd25341`. Soul telemetry now names
`runtime_spine::tests::repo_model_incorporates_pass_and_nonpass_soul_verdicts_causally`,
which proves first valid admission, exact route/request/verdict/Modeling-request
bindings, idempotent replay to the same receipt, final frontier disposition,
and no remaining actionable Hands frontier. The targeted test, launch-context
test, and full library suite all pass; full result is 631 passed, 1 ignored.
The exact `2dd25341` release is published at:
`F:\Projects\Epiphany\.epiphany-run\releases\2dd2534197b38df764b790bd9377374420e01d97\sha256-125e80991b94d8ac06c973bce434c3a996ce93bf44c03505be28e702f6132a0e`
Its witness SHA-256 is
`sha256-6e9fe9fa34fe7b442b8b0f7e9ffdc1f914ebaa5accf32154fc23bcd58dc1891e`.
Soul returned `pass` and the coordinator accepted it. The following exact
release Modeling worker produced the first semantically valid incorporation
result; Mind admitted it immediately as `checkpoint-ready`. The review emitted
`roleAccept:modeling`, no `roleAdmissionRejected`, no retry, and no actionable
Hands frontier.
The v23 local Verse required its `epiphany-local` topology to be reseeded after
the ephemeral launch boundary while retaining the authenticated
`epiphany-starfire` release witness. That split advertisement is now repaired
in the v23 store and should be treated as a lifecycle shakedown finding, not as
permission to bypass release authentication.
The next typed proposal,
`shakedown-v23-eyes-release-topology-evidence-r1`, exposed two additional
authority faults. The exact proposal request is
`repo-frontier-proposal-modeling-febeaa48f17a62ae7d5b8658bf5336890300b40e28c7b0415f62a29a1e43526f`.
First, the coordinator appended verdict-incorporation context before honoring
explicit proposal authority, so historical accepted Soul evidence hijacked the
new launch. Commit `b27b8f41` makes proposal authority exclude that historical
context. Its exact authenticated release
`sha256-72695eb146eb6ad0a8b46a66c7cee181dbdf683a9ad68230d2ac5d32fcb312b1`
was published and live-proven: replaying the same proposal request launched a
fresh Modeling worker and Mind accepted its bounded read-only Eyes frontier.
Second, Self projected only actionable Hands frontiers. The admitted Eyes
frontier therefore fell through to `awaitFrontierProposal`. Commit `d849b17e`
adds an admitted, dependency-ready, challenge-aware Eyes frontier signal and
routes it to Research without granting Hands authority. Focused routing,
runtime-spine eligibility, MVP-status, and full library tests pass; full library
result remains 631 passed, 1 ignored.
The exact `d849b17e` release was published as
`sha256-90e31c86ae45683fc762f8c99e1596eb04fab6e2e39a94beb55a6aa3f1685224`
with witness
`sha256-c560200a10f263e0984759ddc846f3422c951f443422625cb071d6d9791508d8`.
Its live no-override coordinator run still returned `awaitFrontierProposal`.
The canonical item was correctly `recommended_next_organ: Eyes`; Self withheld
it because the newer repository Body challenged its target claim. That rule is
correct for Hands and Imagination, but incoherent for Eyes, whose work is to
investigate challenged claims.
Commit `ea475aac` splits the invariant: Hands and Imagination remain blocked by
challenged targets while Eyes remains routeable. Its exact package was published
as `sha256-873027895a81cec207b23002e3902a796e5d1b7c3a11b7c0195412ece258fe63`
with witness
`sha256-e85d92fd1217e6c9837284919c1cee8e1a6a1c59d82ff7c1df3e8b38a4e3307f`.
The package returned `awaitFrontierProposal`, while the same current-tree status
derivation over v23 reports `Hands=false Eyes=true` and `launchResearch`.
Commit `84f00929` conservatively made every target root source-commit-owned and
added canonical Hands/Eyes readiness to native runtime status. Its exact release
`sha256-dc1967df0a6bed550d7fc8c6eb35853dd8f66431b1413174e5d44535b84b36bc`
launched Research from the challenged frontier. Mind accepted the first packet
as `accept-research-result-worker-70e89ac9-b09d-4de6-8524-25891c13d38b`.
The packet grounds the local Verse topology fail-closed boundary and authenticated
commit `2dd25341` via typed Hands receipt. It honestly leaves the non-bypass
repository-body authentication seam open because the hinted `cultmesh.rs` and
`release.rs` paths do not exist. Modeling must locate the actual current owner.
The run also exposed the actual cause of the earlier package/source decision
mismatch: packaged coordinator status called `native_json("epiphany-mvp-status")`,
but that binary is absent from packaged siblings, so Self silently spawned the
ambient debug executable. Commit `148d4527` replaces this with in-process status
derivation and omits auxiliary heartbeat/Persona/Void projections from routing.
Focused tests and a local executable smoke pass.
Exact `148d4527` was published as
`sha256-39ba935db21602bd52b9f0ba7781b65198ed7ad7d3722d72405c8bebbe1ccc3e`
with witness
`sha256-8fdc1f47995b564d5b48f50d1c0e4d9b8582278b3ce108169477c0a829f8092a`.
Its plan-mode artifact proved heartbeat, Persona, and Void projections are
`omitted`, so packaged Self owns routing status in-process. The Eyes-bound
reorient result cited the accepted evidence and frontier exactly and was accepted
as `accept-reorient-result-worker-c8c27852-71ba-4e65-8a34-68396ff7d5f6`.
After acceptance, Self returned manual regather even though Eyes evidence was
newer than the last Modeling boundary. Commit `4e95c42c` added the missing
single-use route. Its exact packaged replay selected `launchModeling`, then
failed before worker launch because generic Modeling context tried to commit an
older Soul frontier request whose model identity no longer matched v23. That is
an authority collision, not a bad Eyes result.
Commit `6f019cf8` separates the causes: accepted Research newer than the last
Modeling boundary receives an explicit Eyes-to-Modeling handoff and cannot enter
historical Soul-verdict incorporation. The focused context test, all 18
coordinator tests, and all 14 packaged coordinator tests pass.
Exact `6f019cf8` was published as
`sha256-e33317e552aef8ed175c768da96e52952da5962f78b030ced50ad064bf1d8adf`
with witness
`sha256-94bc14b1cb6a6d2cb78cd7c7498964448e80a16f562b341f64175b02b55e40ea`.
Packaged Self launched Modeling as runtime job
`4619ec7a-d9af-4b20-8910-bb5eb371b803` from the accepted Eyes handoff without
touching the obsolete Soul frontier request. Mind accepted the first valid
result as
`accept-modeling-result-worker-4619ec7a-d9af-4b20-8910-bb5eb371b803` at state
revision 17 with evidence
`ev-modeling-eb5aa53a-e679-4993-836a-3ac332b467ef`.
`modelingResultAcceptedAfterResearch` is true, and three later coordinator
steps returned manual regather without relaunching Modeling. The single-use
route is consumed.
The accepted result preserved the remaining honest wound: current source still
does not ground a non-bypass release-authentication enforcement seam. Do not
relaunch the sealed Eyes packet merely to repeat that conclusion.
v22 is a preserved routing-deadlock witness: an explicitly selected corrective
proposal could not preempt the failed verdict-incorporation route. Do not grind
or mutate v22.
Eyes, live packaged Imagination, Persona consequence, Continuity crash/restart/closure,
retention, long-duration resource behavior, and Linux/Yggdrasil cognition are
still unproven.
The local cut now makes proposal admission accept only exact routeable organ
identities `Hands`, `Eyes`, or `Imagination`. Modeling prompt and launch context
state that exact vocabulary. Planning eligibility exposes
`recommendedNextOrganValid` and includes case-mismatched Imagination-shaped
legacy documents as ineligible diagnostics; it does not case-fold them into
authority. Focused hostile-admission, challenge/eligibility, canonical launch
context, and MVP status tests pass.
## Next action
Keep v23 live and v22 sealed as the deadlock witness. The local implementation
now gives packaged Self a read-only typed lifecycle projection plus explicit MVP
actions for planning selection, exclusive Imagination launch, Mind review
request/launch, and atomic decision commit. The full adoption timeline,
Hold/Refuse terminal behavior, worker failure projection, all 19 coordinator
tests, and all 14 MVP coordinator tests pass. Commit and package this pass, then
create legitimate live Imagination pressure through the existing proposal and
proposal-Modeling path. Prove the exact package advances v23 through Imagination
and dedicated Mind without overrides or a second planning substrate.
Commit and push the canonical-organ cut, then package and publish that exact
commit. Cold-start v26. Prove the complete native path without overrides:
Exact `881b2b1a` was packaged and published as
`sha256-e41a428b97a9e67c18425cf0862f15cb9f1de87d03fd15c641dec7f835f0707c`
with witness
`sha256-56ecec1ce5f23256a25dd6d341f6ff73dcf931fcd590de208d3ab3f220bce365`.
Packaged proposal intake created `shakedown-v23-release-auth-seam-r1` and
selection `repo-frontier-proposal-modeling-2a98b0ad0d4f54bc72618ee8b75e20f347bfe3bdf898d3d7247ba9b89bc244c8`.
The first packaged Self pass exposed a surviving split owner: canonical status
returned manual regather, while the CLI could only rewrite
`awaitFrontierProposal`. No worker launched and v23 remained revision 17.
The current local repair makes pending proposal selection a native status
signal, routes it before regather, and demotes the CLI ID to an equality check.
Debug status over live v23 now derives `launchModeling` and that exact request.
Package and publish the repair before retrying; do not use `881b2b1a` to bypass
its own finding.
1. Self derives proposal-bound Modeling.
2. Mind admits one active frontier with sorted source scope and exact
`recommended_next_organ=Imagination`.
3. Self selects and launches Imagination.
4. Imagination emits one typed plan.
5. Dedicated Mind accepts, holds, or refuses it through the typed lifecycle.
6. Inspect the next typed action rather than assuming Hands authority.
Exact `737f2b94` was subsequently packaged and published as
`sha256-b6dfa923a10cab369d56bcbf5d87c2b2da76bc5c3a7f3007ba228f4a63a3564f`
with witness
`sha256-185bd9a3a8657bd9f6cf6a4d158d6368db73363e57ee900ad9cdee265ed6cf82`.
Live v23 proved canonical Self selected `launchModeling` without
`--proposal-modeling-request-id`. The proposal-bound result was not admitted:
it attempted `upsert_frontier` for the already-present
`frontier-eyes-release-topology-evidence-r1`, and Soul superseded the result.
The request is consumed and v23 is revision 19. Do not replay it.
The cause was missing Body context, not weak rejection. Modeling launch context
listed current domains and claims but omitted current frontier items, so the
worker could not know whether a frontier id required insert or revision. The
local repair adds `existingFrontier` to the canonical RepoModel shape and states
the typed contract: upsert only new ids, revise existing ids. The focused
canonical-model launch-context test passes. Commit/package/publish this repair,
then use one fresh legitimate proposal to continue the native lifecycle.
## Immediate re-entry
1. Run `cargo run --manifest-path .\epiphany-core\Cargo.toml --bin epiphany-state -- status`.
2. Read `state/map.yaml` and this handoff.
3. Confirm git HEAD, the exact 6f019cf8 published witness above, and whether the
native Imagination implementation has been committed and packaged.
4. Treat v22 as evidence, not active authority.
Replace this document when state changes. Old attempts belong in evidence and
git, not in the living handoff.
After that proof, continue the actual readiness campaign. Persona public
consequence, Continuity crash/restart/closure, retention bounds, long-duration
resource behavior, and Linux/Yggdrasil cognition remain open. Yggdrasil is the
small canonical public crossing Body; Starfire remains the temporary cognition
and release forge until measured runtime demand justifies resizing Yggdrasil.

Binary file not shown.

View file

@ -34,15 +34,15 @@ invariants:
- Evidence and git retain history; this map retains only state that changes future decisions.
current_status:
summary: Epiphany is a supervised engineering alpha on Starfire, not deployment-ready. Exact 717c649e is authenticated and published as sha256-cec12433. Clean live v24 proved packaged Self derives proposal Modeling without overrides and Mind admits canonical sorted-unique source_scope. It then exposed the next admission/selection split: the worker emitted status proposed, proposal admission accepted it, and Self correctly refused to plan because only active unresolved frontier is routeable. The local repair requires active status at proposal admission, tells Modeling the same contract, and exposes status validity in typed eligibility. Focused rejection, eligibility, launch-context, and MVP status tests pass. v24 is now evidence; a clean next generation must prove Modeling-to-Imagination-to-Mind after this repair is packaged. Persona consequence, Continuity execution, retention, long-duration pacing, and Linux/Yggdrasil cognition remain open.
summary: Epiphany is a supervised engineering alpha on Starfire, not deployment-ready. Exact 872d8ad6 is authenticated and published as sha256-b97f457c with witness sha256-40d6bda4. Clean live v25 proved packaged Self derives proposal Modeling without overrides and Mind admits active canonical sorted-unique source_scope. It then exposed the next admission/selection split: Modeling emitted lowercase recommended_next_organ=imagination, Mind admitted it, and Self's exact canonical Imagination selector correctly found no candidate. The local repair makes exact routeable organ identity a Mind admission invariant, tells Modeling the canonical Hands/Eyes/Imagination vocabulary, and exposes organ validity in typed planning eligibility. Focused hostile-admission, eligibility, launch-context, and MVP status tests pass. v25 is sealed evidence; a clean next generation must prove Modeling-to-Imagination-to-Mind after this repair is packaged. Persona consequence, Continuity execution, retention, long-duration pacing, and Linux/Yggdrasil cognition remain open.
operational_state:
local_worktree: proposal_frontier_active_status_invariant_repair_uncommitted
local_worktree: proposal_frontier_canonical_organ_invariant_repair_uncommitted
operator_wireguard: restored
yggdrasil: canonical_small_public_crossing_body
starfire: temporary_epiphany_cognition_body
starfire_runtime_identity: epiphany-starfire
starfire_swarm_brake: released
starfire_release: authenticated_6f019cf8_published_eyes_modeling_route_proven
starfire_release: authenticated_872d8ad6_published_v25_modeling_route_proven
starfire_repository_body: bound_and_observed
starfire_verse_topology: bound_to_current_repository_body
starfire_native_status: supervised_shakedown_active
@ -52,8 +52,8 @@ current_status:
bifrost_feedback_crossing: authenticated_transport_only
yggdrasil_epiphany_services: absent
resident_physiology: heartbeat_and_self_running
unresolved_authority: v24 revision 3 contains one admitted proposed Imagination frontier whose status makes it permanently ineligible for Self selection. Insert-only proposal authority provides no coherent typed repair, so v24 must not be used to prove native Imagination. The local active-status admission/context/eligibility repair is focused-test-proven but unpublished. Native Imagination live proof, Persona consequence, Continuity crash/restart/closure, retention, long-duration behavior, and Linux/Yggdrasil cognition remain open.
next_action: Commit, push, package, and publish the proposal-frontier active-status invariant repair. Seal v24 as evidence, bootstrap a clean next runtime generation, and prove one exact packaged proposal-Modeling result is admitted active with canonical source_scope, selected by Self, launched through Imagination, and decided by dedicated Mind without overrides.
unresolved_authority: v25 revision 3 contains one admitted active frontier whose lowercase recommended_next_organ makes it permanently invisible to Self's canonical Imagination selector. Insert-only proposal authority provides no coherent typed repair, so v25 must not be used to prove native Imagination. The local canonical-organ admission/context/eligibility repair is focused-test-proven but unpublished. Native Imagination live proof, Persona consequence, Continuity crash/restart/closure, retention, long-duration behavior, and Linux/Yggdrasil cognition remain open.
next_action: Commit, push, package, and publish the proposal-frontier canonical-organ invariant repair. Seal v25 as evidence, bootstrap a clean v26 runtime generation, and prove one exact packaged proposal-Modeling result is admitted active with canonical source_scope and recommended_next_organ=Imagination, selected by Self, launched through Imagination, and decided by dedicated Mind without overrides.
active_subgoals:
- Prove Persona, Self, Eyes, Modeling, Imagination, Hands, Soul, Mind, Continuity, and nervous-system behavior individually.
@ -63,10 +63,10 @@ active_subgoals:
organ_readiness:
persona: partial; authenticated feedback transport is live, but public cognition and speech consequence still need a fresh end-to-end proof.
self: partial; proposal recovery, Mind admission, Hands routing, Soul routing, verdict incorporation, proposal preemption, challenge-aware Eyes routing, in-process status ownership, and single-use Research-to-Modeling routing are live. Native Imagination orchestration is locally implemented and test-proven; packaged live proof is pending.
self: partial; proposal recovery, Mind admission, Hands routing, Soul routing, verdict incorporation, proposal preemption, challenge-aware Eyes routing, in-process status ownership, and single-use Research-to-Modeling routing are live. Exact 872d8ad6 also proved native proposal-to-Modeling launch and active frontier admission; canonical organ identity still blocks packaged Imagination proof.
eyes: proven for one bounded challenged-claim circuit; exact 84f00929 launched Research and Mind accepted the first source-gap packet with typed evidence/checkpoint, exact 148d4527 accepted its causal reorientation, and exact 6f019cf8 proved downstream Modeling consumption.
modeling: proven for the current bounded circuit; proposal-bound source inspection, Mind admission, exact routed identity preservation, and first-valid verdict incorporation completed without retry churn. Broader Body-map coverage remains open.
imagination: unproven live in the current campaign; the typed lifecycle and MVP route are locally implemented, while v23 exposed source_scope routeability and v24 exposed active-status routeability at the proposal admission seam.
imagination: unproven live in the current campaign; the typed lifecycle and MVP route are locally implemented, while v25 exposed that recommended_next_organ identity was admitted as an arbitrary string even though Self routes exact canonical organ names.
hands: proven for bounded consequences; native patch, command, commit, push, and typed receipts bind reviewed path ceilings to exact changed-path subsets.
soul: proven for the current bounded consequence; the exact 2dd25341 release launched Soul, produced an accepted pass verdict grounded in focused and full-library receipts, and routed verdict incorporation.
mind: partial; RepoModel admission, thread-state admission, gateway review, state commit, and first-valid verdict incorporation are live; broader stale or overstated frontier correction remains open.