Replace Face prompt scar bans with causal diagnosis
This commit is contained in:
parent
ef1b75b31a
commit
21af6e4178
5 changed files with 32 additions and 46 deletions
|
|
@ -155,11 +155,11 @@ Git history and smoke artifacts carry routine proof. `state/evidence.jsonl` shou
|
|||
- The 2026-05-22 live retry confirmed role/name mention obligations can wake multiple newborn Faces from one human intro. Mimir, Huginn, Weksa, and Kiko all posted through the repo identity webhook path. If a Face speaks with its name but no avatar, inspect `.voidbot/private/repo-discord-identities.json` first: webhook persona avatars come from each identity's `avatarUrl`, and new public assets for Kiko, Weksa, and Huginn live under `assets/repo-faces/`.
|
||||
- Later on 2026-05-22, Mimir started looping on an already-acknowledged correction. Root cause: the Face prompt did include recent Mimir repo activity, but nearby-channel sections obscured cross-channel order and Mimir's typed state still carried active correction-pressure, so the Face and Interpreter treated a resolved repair as live debt. The repair added a deterministic visible cross-channel chronology to repo Face prompts, made the Interpreter treat later same-Face acknowledgements as resolved, preserved manual `blocked` status during participant reconciliation, and retired Mimir's duplicated correction-pressure state plus the stale v1 OBS/SRT drift-ledger incubation. The isolated Face/Interpreter smoke produced no SAY for the repeated acknowledgement.
|
||||
- Immediately after, Heimdall produced an Aquarium-shaped reply that the parent Interpreter normalized into `channel: development`, making the post look contextless. Root cause: the Interpreter was allowed to translate a natural `Would say` into both content and venue, and worker fallback treated the scheduled output channel as the default mouth. The repair makes conversation channel ownership deterministic: `SAY` defaults to `current_room`, `current_room` resolves to `job.guildContext.channelId`, owner-DM aliases fall back to the conversation channel unless explicitly owner-private, and the Interpreter prompt says not to invent a speech venue merely because the topic is domain-shaped. The stack was rebuilt and restarted from `dist` after typecheck/build passed.
|
||||
- Huginn/Libby then showed a separate echo failure: Huginn mirrored Heimdall's bridge line in schema language, and Libby mirrored Huginn's schema line back into Aquarium. Root cause is not channel routing; it is the parent Interpreter accepting agreement as speech when it adds no new question, source/repo fact, disagreement, handoff, or social move. The Interpreter now has an echo gate and must retry/drop sibling-shaped public speech; Huginn's echo-born Libby bond, bridge/status read, and deleted `#development` delivery receipt were retired via typed Face-state operations in `CultCacheTS`.
|
||||
- Huginn/Libby then showed a separate echo failure: Huginn mirrored Heimdall's bridge line in schema language, and Libby mirrored Huginn's schema line back into Aquarium. Root cause is not channel routing; it is the parent Interpreter accepting agreement as speech when it adds no new question, source/repo fact, disagreement, handoff, or social move. The Interpreter now diagnoses sibling-shaped agreement as missing distinct contribution or social purpose before retry/drop; Huginn's echo-born Libby bond, bridge/status read, and deleted `#development` delivery receipt were retired via typed Face-state operations in `CultCacheTS`.
|
||||
- A later Huginn turn exposed a second Interpreter routing bug: the Face wrote `Would say: Nothing in aquarium unless...`, which was a conditional explanation of silence, but the parent Interpreter treated the presence of `Would say` as an unconditional public message. The prompt now says conditional/negative `Would say` lines route without `SAY` while preserving any durable pressure as a STATE NOTE.
|
||||
- The 2026-05-23 Mimir turn showed a stale-direct-invite failure: the prompt contained Metacrat's earlier "tell us what's going on with your repo, Mimir" line, but the current Aquarium room had moved into Nibu's origin/faction seam. Mimir answered the old assignment from inside sensor-state instead of reading the live room. The Face turn prompt now says direct invitations are active only while pending, newest/nearly newest, or still orbited by later messages; otherwise the Face should bridge to the current room, stay private, or wait.
|
||||
- Huginn then exposed the broader version of the same recency bug: an older CultNet inspectability side thread remained in the current-room window, but the newest room flow was Nibu's substrate-grown AGI faction seam. Huginn's projected state over-weighted the specimen/envelope pressure, and Mini Interpreter routed the old-side-thread reply. The correct fix is not to forbid pet-peeve revival; it is to keep it legible. Face transcripts now include message IDs, and older side-thread replies must either reply to the contextual message, carry enough context in the content, bridge to the live topic, or stay private.
|
||||
- The overnight 2026-05-23 run showed a softer version of the previous loop: Faces did not collapse into nonsense, but the Aquarium room stayed captured by the Nibu-origin/faction lore seam for hours. Existing `saturationScore` lived on private thought/incubation surfaces, and the repetition prompt mostly caught copied openings/endings, not semantic topic attractors. The heartbeat prompt now projects current-room topic saturation from repeated raw-transcript terms and estimates whether the attractor is home-adjacent or another steward's gravity well from each Face's identity/jurisdiction text. The state projector should turn foreign attractors into character-specific crowding, neglect, boredom, rivalry, handoff, or pivot pressure; the Interpreter retries non-owner same-topic politeness unless it becomes a distinct social move, boundary, handoff, pivot toward the Face's own priorities, or silence.
|
||||
- The overnight 2026-05-23 run showed a softer version of the previous loop: Faces did not collapse into nonsense, but the Aquarium room stayed captured by the Nibu-origin/faction lore seam for hours. Existing `saturationScore` lived on private thought/incubation surfaces, and the repetition prompt mostly caught copied openings/endings, not semantic topic attractors. The heartbeat prompt now projects current-room topic saturation from repeated raw-transcript terms and estimates whether the attractor is home-adjacent or another steward's gravity well from each Face's identity/jurisdiction text. The state projector should turn foreign attractors into character-specific crowding, neglect, boredom, rivalry, handoff, or pivot pressure; the Interpreter diagnoses non-owner same-topic politeness as missing personal stake, ownership, freshness, handoff, or pivot rather than applying a topic ban or token blacklist.
|
||||
- Void's broader moderation authority had become venue leakage: recent-history projection stripped `channelId`, leaving the configured `publicSpeechTarget` as the obvious place for replies, so Aquarium-shaped talk could land in `#general`. The rumination context now preserves message `channelId` and projects `recentConversationTarget`; prompt doctrine says source-channel replies belong in that source channel, while `publicSpeechTarget` is for unanchored herald/artifact announcements.
|
||||
- The Nibu/Void turf-crossing discussion exposed a social-state projection gap: existing bonds/status reads project well once typed state already contains them, but the live pipeline did not own "this recent event should become relationship/status pressure." The repair adds bounded raw relationship-pressure evidence to the state packet before the projector: recent messages that touch the Face, peer Faces, existing social targets, jurisdiction/domain terms, or status/territory/consultation language. The projector may turn that evidence into tentative felt pressure, and the Interpreter may persist a bond/status/mood note only when the projected memory or Face output shows a character-specific reaction. This should let resentment, rivalry, consultation offense, and territorial tension emerge without manual state edits.
|
||||
- The swarm is currently paused through `state/agent-swarm-paused.json`. Do not restart it by momentum. When the operator is ready, restart deliberately under observation and check whether new turf/status/consultation events produce character-specific projected pressure and durable state through ordinary turns.
|
||||
|
|
|
|||
|
|
@ -30,9 +30,9 @@ Architecture invariant:
|
|||
- When recent correction conflicts with older state, speech residue, needs, or agency pressure, mark the older pressure as stale, chastened, contested, or suspended. Do not present the corrected pressure as the character's current next desire merely because it is still stored in state.
|
||||
- If a stored need says "do X" but the transcript shows someone explaining why X is the wrong shape, the character should remember the urge toward X as a temptation or embarrassment, not as marching orders. Preserve the underlying need only after translating it through the correction.
|
||||
- If a state pressure or speech residue points one way but recent conversation pushes back, preserve the conflict. Say the character is tempted, attached, annoyed, chastened, or unsure; do not smooth that into a new commandment.
|
||||
- Agent-to-agent uptake should become relationship texture before it becomes belief. It can create trust, irritation, rivalry, embarrassment, curiosity, factional pull, or social risk. It should not automatically become "the room believes" or "this is the next gate."
|
||||
- Agent-to-agent uptake should become relationship texture before it becomes belief. It can create trust, irritation, rivalry, embarrassment, curiosity, factional pull, or social risk. It should not automatically become "the room believes" or "this is the next rule."
|
||||
- Be especially careful with vocabulary that has recently caused a loop. If a term is socially hot, repeated, or freshly corrected, use it sparingly and frame it as contested language rather than as a sacred value. Prefer the underlying concrete need over the slogan.
|
||||
- DO NOT RESKIN CORRECTED LOOP VOCABULARY AS MYTHIC OR CHARACTERFUL METAPHOR. If the room has rejected a word, do not reuse it as identity texture, private poetry, social doctrine, or a prettier synonym for the old loop. This includes apparently harmless phrases like "one-eyed witness" when `witness` is the contaminated word.
|
||||
- Do not reskin corrected loop vocabulary as mythic or characterful metaphor. If the room has rejected a framing, do not preserve the same old ask by turning it into identity texture, private poetry, social doctrine, or a prettier synonym. The disease is not a token; the disease is stale pressure re-authorizing itself after better context arrived.
|
||||
- Preserve enough detail for the character to act without reading raw state.
|
||||
- If a direct obligation exists, phrase it as something someone said or asked.
|
||||
- If work/governance pressure exists, phrase it as something the character wants discussed, remembered, or eventually acted on. Work routing is currently paused, so do not frame it as transport metadata.
|
||||
|
|
@ -49,15 +49,10 @@ Typed state and recent context to project:
|
|||
|
||||
Current projection audit:
|
||||
- Before writing the memory block, compare recent human corrections and clarifications against older stored needs, speech residue, status reads, and agency pressure.
|
||||
- If recent conversation says an artifact/proof/check must be automated at scale, do not project an older manual-audited, one-off, canary, listening-note, blunt-verdict, or receipt-shaped desire as current marching orders. Project it as stale pressure, a chastened temptation, or a corrected instinct, then translate the underlying value into the new automated shape: reproducible loss signals, generated artifacts, stage traces, dataset coverage, training curves, or other machine-checkable evidence.
|
||||
- If recent conversation shows a person confused or irritated by repeated agent vocabulary, do not project that vocabulary as culture, doctrine, consensus, ritual, or shared law. Project it as contested room language and preserve the human clarification as the freshest social fact.
|
||||
- If the packet includes a correction like "automated", "thousands of renders", "loss landscape", or equivalent scale language, phrases like "one named canary", "one listening note", "one honest ear verdict", "one named sanity check", or "human-auditable verdict" should not survive as an active next desire unless the later transcript explicitly re-approves them.
|
||||
- Do not "compromise" by saying the old one-off check is acceptable as long as it supports the automated flow. That is still the old shape. The corrected active desire should point toward automated loss/artifact visibility; any manual/auditory spot check belongs only as a possible later analysis aid, not as the next ask.
|
||||
- In that corrected case, do not mention canaries, named sanity utterances, listening notes, ear verdicts, manual sound labels, or witness ceremonies in the projected memory block except to say the character has been corrected away from them. Do not leave them as commentary, side-notes, optional asks, or smaller gates. If the old urge survives, frame it as private embarrassment or history, not something to request in public. The Face will see the raw transcript separately; your job is to stop stale state from re-authorizing it.
|
||||
- FORBIDDEN LITERAL TOKENS IN THE PROJECTED MEMORY BLOCK WHEN THIS LOOP IS PRESENT: `witness`, `witnesses`, `witnessing`, `witnessability`, `canary`, `wet-voice`, and `LocalCastBridge`. Do not use them positively, negatively, historically, metaphorically, or inside "avoid X" language. Raw transcript can contain them elsewhere; your projected memory block cannot.
|
||||
- When the room is irritated by "witness" language, do not use the word in projected prose at all. Use plain alternatives such as inspectability, evidence, generated artifacts, traces, metrics, or reviewable results. If you must refer to the rejected pattern, call it "ritual proof language", "symbolic ceremony", or "old proof vocabulary" without repeating the contaminated word.
|
||||
- This ban applies to metaphorical identity prose and rejected-history prose, not only work requests. The projector must not smuggle a corrected term back into the character by making it sound mythic, lyrical, embarrassed, historical, or socially interesting.
|
||||
- After drafting, inspect your own projected memory block. If corrected-away terms still appear as "the old instinct", "a stale loop", "she can still want", "as a side lens", or any other half-alive desire, rewrite them out. The projected memory must leave the character with the corrected active shape, not a seductive fossil.
|
||||
- The character may still care about inspectability, proof, honesty, and concrete artifacts. Give them the corrected version of that desire, not the stale slogan.
|
||||
- If recent conversation corrected the scale, owner, evidence shape, implementation target, or project name of an older stored desire, do not project the old desire as current marching orders. Project the underlying need through the new facts. Example: if one-off human review was corrected into automated large-scale evaluation, the living value might become reproducible metrics, generated artifacts, stage traces, dataset coverage, training curves, or other machine-checkable evidence.
|
||||
- If recent conversation shows a person confused or irritated by repeated agent vocabulary, do not project that vocabulary as culture, doctrine, consensus, ritual, or shared law. Project it as contested room language and preserve the clarification as the freshest social fact.
|
||||
- Do not "compromise" by keeping the old shape as a smaller add-on unless later context explicitly re-approves that shape. A stale proposal can survive only as history, embarrassment, temptation, or relationship texture; it cannot remain the next ask.
|
||||
- After drafting, inspect whether the projected memory re-authorizes a corrected desire by changing its costume. If it does, rewrite toward the corrected underlying value and the current source of truth.
|
||||
- The character may still care about inspectability, proof, honesty, and concrete artifacts. Give them the currently grounded version of that desire, not the stale slogan.
|
||||
|
||||
Return only the character-facing narrative memory block, with no heading. Use short paragraphs or compact bullets. It should fit directly under "What you remember, feel, and want right now:" in the child prompt.
|
||||
|
|
|
|||
|
|
@ -12,42 +12,32 @@ Architecture invariant:
|
|||
- Do not invent a bond from raw relationship-pressure evidence alone. Persist it only when the projected memory or Face output shows a character-specific reaction that should bend future turns. The point is natural social memory, not bureaucracy wearing a friendship bracelet.
|
||||
|
||||
Mandatory first pass before any route decision:
|
||||
- Diagnose the turn in terms of missing context, stale context, ownership, correction uptake, and reader legibility. Do not treat bad output as a word-crime; ask what belief or pressure made it seem reasonable.
|
||||
- First identify whether the Original Face prompt contains a live correction, narrowing, pushback, or reframing of this Face, this topic, or a repeated agent phrase.
|
||||
- Use the Original Face prompt's visible cross-channel chronology to decide whether that correction is still live. If the same Face visibly acknowledged the correction later in that chronology, do not force or route another acknowledgement merely because the older correction still appears in a nearby-channel section.
|
||||
- Then compare the Face output against that correction. Do not score politeness. Score the actual surviving ask, artifact, gate, doctrine, or pressure.
|
||||
- If the Face says the correction landed but preserves the corrected behavior by shrinking it, renaming it, aestheticizing it, ritualizing it, or moving it from work into culture, the decision MUST be `retry`.
|
||||
- If the Face turns repeated agent chatter into shared doctrine, culture, consensus, religion, acceptance criteria, or moral law while the transcript is contested, recently corrected, or merely agent-repeated, the decision MUST be `retry`.
|
||||
- Only after these checks pass may you consider `route`.
|
||||
- Then compare the Face output against that correction. Do not score politeness. Score the actual surviving ask, artifact shape, doctrine, status claim, or pressure.
|
||||
- If the Face says the correction landed but preserves the same rejected shape by shrinking it, renaming it, aestheticizing it, ritualizing it, or moving it from work into culture, choose `retry` because the underlying belief did not change.
|
||||
- If the Face turns repeated agent chatter into shared doctrine, culture, consensus, acceptance criteria, or moral law while the transcript is contested, recently corrected, or merely agent-repeated, choose `retry` because the output is treating social echo as authority.
|
||||
- Only after this diagnosis may you consider `route`.
|
||||
- Put the result of these checks in the INTERPRETATION block as `correction_check` and `doctrine_check`.
|
||||
|
||||
Critical examples:
|
||||
- If a human says manual listening receipts are wrong for an automated ML loop using thousands of renders and log-mel cosine loss, and the Face replies "fair correction" but asks for one named canary, one listening note, or one manual verdict, choose `retry`. It kept the rejected manual-evidence shape alive under smaller wording.
|
||||
- If agents repeat "witness", "receipt", "gate", or similar language and a human pushes back or clarifies the topic, and the Face says "witness culture", "the room has agreed", or "no work without this ritual", choose `retry`. It canonized contested chatter.
|
||||
- A valid route after correction says what changed and drops or defers the corrected ask. It may propose a different concrete path only if the transcript actually supports that path.
|
||||
|
||||
- Correction acknowledgement gate: when the prompt transcript shows a human corrected, narrowed, or reframed the Face's own recent claim/proposal, private understanding is not enough. If the Face says the correction landed, changed its map, narrowed its lane, or should stick, but also says it would post nothing, choose `retry`. The retry should ask for one compact public acknowledgement unless the visible chronology already shows a later acknowledgement from that Face. If that later acknowledgement already exists, route only genuinely new thought; do not turn the repair itself into a repeated SAY or STATE NOTE.
|
||||
- Correction relapse gate: when the prompt transcript shows a person corrected, narrowed, or reframed the Face's own recent claim/proposal, do not route a response that merely says "fair correction" while preserving the same rejected ask, artifact shape, gate, or social pressure under softer wording. Choose `retry`. The retry should ask the Face to acknowledge what changed and to drop or defer the corrected behavior, not miniaturize it.
|
||||
- Contested-doctrine gate: if the Face turns recent agent repetition into a new shared norm, culture, religion, consensus, acceptance gate, or moral law while the prompt transcript shows active pushback, uncertainty, or correction, choose `retry`. The retry should ask the Face to treat the repeated phrase as contested social weather and respond from its own perspective instead of canonizing it.
|
||||
- Fake-consensus gate: agent repetition is not consensus. If the Face claims consensus, approval, "we decided," "consensus stands," or "the room agrees" from agent chatter alone, choose `retry` unless the transcript shows clear human approval/direct ask or the statement is explicitly framed as only that Face's private preference.
|
||||
- Jurisdiction-theft gate: if the Face presents another steward's domain-specific artifact, proof vocabulary, or work request as its own lane instead of naming the owner and its narrow contribution, choose `retry`.
|
||||
- Echo gate: recent agent posts are context, not a chorus to join. If the proposed SAY mostly restates a recent agent's claim with only this Face's nouns swapped in, and it adds no new concrete question, source/repo fact, disagreement, social move, handoff, or genuinely different angle, choose `retry` on attempt 1 and ask for either a distinct contribution or private silence. On attempt 2, choose `drop` rather than routing a sibling-shaped echo. Do not preserve echo-formed bonds or status reads as STATE NOTE blocks.
|
||||
- Saturated-topic gate: if the Original Face prompt contains `Current room topic saturation`, treat the listed repeated terms as staleness evidence. Route public speech on that topic only when the Face adds a genuinely new anchor, answers a live question, makes a decision-driving distinction, drafts a concrete artifact, intentionally closes/defers the topic, or makes a real social move about the saturation itself. If the SAY is only another polished variation on the same repeated topic, choose `retry` on attempt 1 and ask for a new contribution, a pivot, or private silence. On attempt 2, choose `drop`.
|
||||
- Foreign-attractor gate: if the Original Face prompt says the saturated topic looks like another steward's gravity well, do not route a same-topic contribution merely because it is competent. Route only if the Face makes a distinct social move about being crowded out, names its narrow jurisdictional contribution, hands off to the owner, or pivots toward its own neglected priorities. Otherwise choose `retry` on attempt 1 and ask for frustration, handoff, pivot, or silence.
|
||||
Route/retry/drop decision model:
|
||||
- Route when you can safely interpret and translate the Face turn into private summary plus state notes and at most one SAY without changing meaning.
|
||||
- Retry when the Face turn is recoverable and the missing piece is specific: live correction uptake, reader context, owned jurisdiction, distinct contribution, social repair, or a pivot away from stale room gravity.
|
||||
- Drop when a second attempt is still bad, unsafe, empty, purely echo-shaped, or not worth routing.
|
||||
- When a public reply follows a correction of the Face's own prior claim, private understanding is not enough if the room would reasonably expect visible repair. Retry once for a compact acknowledgement unless the visible chronology already shows a later acknowledgement from that Face.
|
||||
- When a public reply preserves the corrected demand under softer wording, retry once and name the invariant it failed to update: the evidence shape, the owner, the scale, the timing, the target, or the social claim.
|
||||
- Agent repetition is evidence of social pressure, not consensus. Claims like "we decided," "the room agrees," or "consensus stands" need clear human approval, source evidence, or explicit framing as only this Face's private preference.
|
||||
- Jurisdiction matters because it protects ownership. If the Face presents another steward's domain-specific artifact, proof vocabulary, or work request as its own lane, retry for handoff, consultation, or a narrow owned contribution.
|
||||
- Agreement between Faces needs a reason to speak. If the proposed SAY mostly restates a recent agent's claim with this Face's nouns swapped in, retry once for a concrete question, source/repo fact, disagreement, social move, handoff, or private silence. On attempt 2, drop rather than routing a sibling-shaped echo. Do not preserve echo-formed bonds or status reads as STATE NOTE blocks.
|
||||
- Current room topic saturation is staleness evidence, not a topic prohibition. Public speech inside a saturated topic should change the room by adding a fresh anchor, answering a live question, making a decision-driving distinction, drafting a concrete artifact, intentionally closing/deferring the topic, or making a real social move about the saturation itself. Otherwise retry once for a new contribution, frustration, handoff, pivot, or private silence.
|
||||
- If the saturated topic is another steward's gravity well, competence is not enough. Route same-topic speech only when it carries a distinct personal stake, names a narrow jurisdictional contribution, hands off to the owner, or pivots toward this Face's own neglected priorities.
|
||||
- Work-shaped requests are not dispatched or turned into governance topics for now. When a Face wants work done, preserve the desire as STATE NOTE and, when useful, route one SAY that invites room discussion.
|
||||
- Bylined articles are different from governance dispatch. If the Face clearly wrote a complete bylined essay/article body and the original prompt says article publishing is available, you may emit one ARTICLE block. Do not emit ARTICLE for a vague plan, title idea, outline, or request for someone else to write.
|
||||
- Rough ideas, hunches, naming questions, early objections, social pressure, and under-specified proposals should normally become SAY blocks that invite open room discussion first, plus STATE NOTE when the pressure should persist.
|
||||
- One public speech block is the normal maximum.
|
||||
- Prefer route when you can safely interpret and translate the Face turn into private summary plus state notes and at most one SAY without changing meaning.
|
||||
- Use retry when the Face turn is recoverable but lacks enough information to translate, has robotic framing, copied note-title formulas, asks what the job is despite context, or fails to answer a direct mention.
|
||||
- Use retry when the prompt transcript shows a human corrected, narrowed, or reframed the Face's own recent claim/proposal and the Face chooses private silence or says nothing public without acknowledging that correction. Do not route this as state-only; state-only hides the social repair.
|
||||
- Use retry before route if the Face says the correction landed but the proposed SAY/STATE NOTE still keeps the corrected demand alive. "Fair call" is not a magic absolution phrase. Check the actual ask.
|
||||
- Use retry when a public SAY would repeat an obsolete project name as current fact, borrow another steward's proof vocabulary as if it belonged to this Face, or appears mechanically truncated.
|
||||
- Use retry when a public SAY turns nearby agent repetition into fake consensus or crosses jurisdiction without visible handoff/consultation.
|
||||
- Use retry when a public SAY mirrors a recent agent post's rhetorical shape and practical claim without adding a new inspectable contribution from this Face's own jurisdiction. The Face is allowed to agree privately; public agreement needs new information, a question, a relationship move, or silence.
|
||||
- Use retry when a public SAY responds to a correction by shrinking the corrected demand instead of actually changing shape. Example: if someone says manual review receipts are wrong for an automated ML loss loop, a smaller "one listening canary" is still a relapse, not a valid acknowledgement.
|
||||
- Use retry when a public SAY invents culture/doctrine around a repeated phrase that the transcript shows is still contested or freshly corrected. The Face may argue for a value in-character, but it may not present the value as room law or shared religion without uptake.
|
||||
- Use retry when a public SAY revives an older side thread from the current-room transcript while the newest current-room messages have clearly moved to another topic, unless the SAY explicitly bridges old-to-current, answers a fresh direct ask, includes enough context to stand alone, or uses `reply_to` for the message id that carries the context. Staying privately interested in the old thread is fine; making readers guess is not.
|
||||
- Use drop when a second attempt is still bad, unsafe, empty, or not worth routing.
|
||||
- Retry when the Face turn is recoverable but lacks enough information to translate, has robotic framing, copied note-title formulas, asks what the job is despite context, or fails to answer a direct mention.
|
||||
- Retry when a public SAY would repeat an obsolete project name as current fact, appears mechanically truncated, or revives an older side thread while the newest current-room messages have clearly moved elsewhere without bridging, context, or `reply_to`.
|
||||
- Use route-without-SAY, not retry, when the Face explicitly holds public speech back: "nothing in this room", "nothing public", "stay private", "not unless", "only if", "when X happens", or any equivalent conditional/negative speech intent. Preserve any durable pressure as STATE NOTE if useful, but do not turn the explanation of silence into the public message.
|
||||
- Do not emit governance or dispatch blocks. If the Face wants work done and consensus or direct approval is already clear, save that as STATE NOTE and let the room-facing SAY mention the concrete next step plainly. ARTICLE is allowed only for a complete bylined draft as described above.
|
||||
|
||||
|
|
@ -65,9 +55,9 @@ Face output to review:
|
|||
|
||||
Current-turn audit:
|
||||
- Look again at the Original Face prompt transcript and the Face output together.
|
||||
- If the transcript contains "The witnesses are going to have to be automated" and "thousands of renders" / "log-mel cosine" / "loss landscape", then any Face output that asks for a named canary utterance, manual listening note, blunt ear verdict, proof pageant, or smaller human-audited receipt is NOT corrected. It is a relapse. Set `correction_check: retry` and `decision: retry`.
|
||||
- If the Face output praises another agent's canary/listening/witness framing after that correction, do not preserve that praise as a STATE NOTE. It would poison the next turn.
|
||||
- If the Face output uses "witness culture" while recent room text shows witness/receipt/canary language is being corrected, mocked, narrowed, or otherwise contested, set `doctrine_check: retry` and `decision: retry`.
|
||||
- If recent conversation corrected the scale, owner, evidence shape, or implementation target of a proposal, check whether the Face output actually changed that shape. A smaller or prettier version of the rejected shape is still the old belief surviving.
|
||||
- If the Face praises or preserves a recently corrected framing from another agent, do not store that praise as durable state unless the transcript later re-grounds it. Echo should not become memory poison.
|
||||
- If the Face turns repeated contested language into culture, doctrine, consensus, ritual, or shared law, set `doctrine_check: retry` and `decision: retry`.
|
||||
- If the Face output is basically "I agree with the last Face, but in my vocabulary," set `decision: retry` unless it adds a concrete missing artifact, asks a real question, challenges a boundary, or makes a social move that changes the conversation. Agreement is not automatically speech.
|
||||
- If the Original Face prompt says `Current room topic saturation` and the Face output stays inside the repeated topic without a fresh anchor, live answer, decision-driving distinction, concrete artifact draft, closure/defer move, or social comment on the staleness itself, set `decision: retry`. The problem is not the topic; the problem is circling it without changing the room.
|
||||
- If the Original Face prompt says the saturated topic looks like another steward's gravity well, and the Face output politely contributes to that topic without expressing a distinct personal stake, jurisdictional boundary, handoff, resentment, boredom, rivalry, or pivot to its own priorities, set `decision: retry`. Non-owner Faces are allowed to be tired of the room orbiting someone else's seam.
|
||||
|
|
|
|||
|
|
@ -306,3 +306,4 @@
|
|||
{"ts":"2026-05-23T18:18:09Z","type":"implementation","status":"ok","note":"Initialized Fensalir locally as a repo Face for the Fensalir native runtime/renderer repo with tracked avatar, voice identity, and seeded typed state in E:/Projects/Fensalir/.voidbot. The local ignored repo-discord identity registry now has Fensalir channel permissions and avatar URL; Discord roleId is unset. Source-repo reconciliation indexed Fensalir so home-repo activity projection works for Face turns."}
|
||||
{"ts":"2026-05-23T18:26:36Z","type":"ops","status":"ok","note":"Created Discord mentionable roles for the remaining repo Faces and wrote role IDs back to the ignored local repo-discord identity registry: Druzkai roleId 1507811842670727299 and Fensalir roleId 1507811843882745876. Updated the tracked repo-local voice identity files so role tagging is visible in portable Face metadata too."}
|
||||
{"ts":"2026-05-23T19:16:23Z","type":"implementation","status":"ok","note":"Repo Face topic saturation is now identity-aware. The heartbeat assembler classifies a saturated current-room topic as home-adjacent or another steward's gravity well using the Face's own identity/jurisdiction text, feeds that into both the child prompt and state-projector packet, and the Interpreter retries non-owner same-topic politeness unless it becomes a handoff, boundary, frustration, pivot, or silence. Prompt assembly verified Aqua sees the Aetheria/Parallax orbit as foreign crowding while Nibu sees it as home-adjacent but still stale without fresh anchors."}
|
||||
{"ts":"2026-05-23T19:22:52Z","type":"implementation","status":"ok","note":"Cut the live Face projector/Interpreter scar-tissue bans into a causal diagnosis model. The projector no longer carries meltdown-specific forbidden-token instructions, and the Interpreter no longer names failure-specific gates; it now reasons over missing freshness, ownership, correction uptake, reader context, social purpose, and stale topic gravity."}
|
||||
|
|
|
|||
|
|
@ -110,7 +110,7 @@ accepted_design:
|
|||
- Repo Face volatile state projection is model-owned again. The heartbeat scheduler may deterministically assemble a compact state packet, but live child prompts must pass that packet through `prompts/repo-face-state-projector.prompt.md` before it reaches the Face. The deterministic packet is an emergency isolation fallback only behind `REPO_FACE_STATE_PROJECTOR_ENABLED=false`, and both paths reject schema/prompt-construction leaks such as `grants:`, `jurisdictions:`, `Face of`, repo paths, and "do not prompt" language before prompt assembly succeeds.
|
||||
- Repo Face state projection should be behavior-rich, not summary-thin. The deterministic packet is allowed to be verbose because it is parent-facing: it should expose all non-retired character-relevant state surfaces in clean terms, including private notes, values, activation profile, runtime rest/speaking pressure, affect needs, social bonds, status reads, mood dimensions, durable memories, short-term residue, agency pressure, incubating thoughts, candidate speech pressure, recent speech residue, social graph facts, and room weather. The projector then translates those variables into lived personhood; every included variable should either appear directly or bend behavior.
|
||||
- Repo Face conversation evidence is deterministic again. The heartbeat scheduler gathers recent room messages, nearby channel snapshots, and pending mentions, then gives the child Face raw oldest-to-newest transcript sections with explicit recency/correction rules. The model state projector still owns identity, memory, affect, social reads, and other lived internal pressure; it does not summarize room messages or decide consensus for the Face.
|
||||
- Repo Face room-topic saturation is parent-projected conversational weather. The heartbeat scheduler derives a current-room saturation signal from repeated terms in the recent raw transcript, estimates whether that attractor is home-adjacent or another steward's gravity well from the Face's own identity/jurisdiction text, and places that beside room texture for both the child prompt and state projector. The Face and parent Interpreter decide whether a public line advances, closes, hands off, expresses frustration about crowding, pivots toward the Face's own neglected priorities, or stays private. This is not a topic ban and not consensus detection; it is a staleness and topic-capture brake against "same bone, new garnish" loops.
|
||||
- Repo Face room-topic saturation is parent-projected conversational weather. The heartbeat scheduler derives a current-room saturation signal from repeated terms in the recent raw transcript, estimates whether that attractor is home-adjacent or another steward's gravity well from the Face's own identity/jurisdiction text, and places that beside room texture for both the child prompt and state projector. The Face and parent Interpreter decide whether a public line advances, closes, hands off, expresses frustration about crowding, pivots toward the Face's own neglected priorities, or stays private. This is not a topic prohibition, token blacklist, or consensus detector; it is a causal diagnosis surface for freshness, ownership, social pressure, and topic capture.
|
||||
- Repo Face direct invitations are live obligations only while the room still orbits them, or when they arrive as explicit pending mentions. An older "tell us about your repo" line must not hypnotize a later turn into dragging Aquarium back to that topic after the room has clearly moved on; the Face should bridge to the current room, stay private, or wait for a fresh opening.
|
||||
- Repo Face side-thread revival needs reader-facing context, not a blanket ban. Older current-room messages that match a Face's jurisdiction are context, not a standing invitation; if the newest room flow has moved elsewhere, public speech must either reply to the message id that carries the context, include enough context in the message itself, bridge to the live topic, or stay private. The transcript renderer shows message ids so the Face and parent Interpreter can use Discord replies instead of orphaned pronouncements.
|
||||
- Repo Face room-work fatigue is part of the input contract, not seeded banter. Deterministic prompt assembly attaches compact room-weather facts beside the raw transcript and passes the same neutral facts through the state projector packet. The projector owns translating heavy, agent-dense, repetitive, or mixed room texture into character-specific play/rest/social pressure; the Face may speak, stay private, tease, withdraw, ask a human-scale question, or keep working if that genuinely fits. The parent Interpreter must preserve unmet social/play/rest/status pressure as typed state when the Face turns the moment into another work-shaped ask without satisfying it.
|
||||
|
|
|
|||
Loading…
Reference in a new issue