diff --git a/wiki/log/2026-07-27-gaslighting-selftrust-capture.md b/wiki/log/2026-07-27-gaslighting-selftrust-capture.md index 6c1a5c29..02213535 100644 --- a/wiki/log/2026-07-27-gaslighting-selftrust-capture.md +++ b/wiki/log/2026-07-27-gaslighting-selftrust-capture.md @@ -14,11 +14,13 @@ in their own memory. Cameron adopted the extension. Captured: - [plots.md](../mechanics/plots.md) REOPENED with criterion 12: an - `erode-self-trust` category, a `tamper` world act (rewrite or erase one - target-authored/witnessed record, signature derived from the record's - channel, persistent discoverable trace), a bounded `self_trust` ending - consequence, and a mandatory corroboration blowback ending. The Marcus bar - is unchanged — no asserted psychic damage. + `erode-self-trust` category, a `tamper` world act that changes the + world-facing artifact while preserving canonical history and acquired + evidence, exact persisted original/altered custody with carrier-derived + signature, a bounded encounter-gated `self_trust` ending consequence, and a + mandatory corroboration blowback ending. The Marcus bar is unchanged — no + asserted psychic damage at plot completion before the target experiences + the contradiction. - [social.md](../mechanics/social.md) declares the self-trust axis on every person (adopted, pending criterion 9): a shaken person is a worse witness and a worse tool; corroboration and tamper-trace discovery restore the diff --git a/wiki/log/decisions/2026-07-27.md b/wiki/log/decisions/2026-07-27.md index fd939df4..0574ec3b 100644 --- a/wiki/log/decisions/2026-07-27.md +++ b/wiki/log/decisions/2026-07-27.md @@ -140,12 +140,19 @@ Owners: [aggregate-observer.md](../../mechanics/aggregate-observer.md) and person's trust in their own perception. - Gaslighting enters as authored plot content, not a new verb. Three vocabulary pieces: a `self_trust` axis on every person (social.md), an - `erode-self-trust` plot category, and a `tamper` world act that rewrites or - erases one record the bound target authored or witnessed so surviving - evidence contradicts their memory (plots.md). + `erode-self-trust` plot category, and a `tamper` world act that changes or + suppresses the world-facing artifact for one record the bound target + authored or witnessed so that artifact contradicts their memory (plots.md). +- Existing exact-custody law constrains tamper: the canonical event, original + payload, and acquired evidence remain immutable. Tamper authors separate + persisted custody for the original and altered artifact, acting persona, + tick, carrier/channel, derived signature, and discovery state. - The Marcus bar and typed-consequence discipline apply unchanged: erosion happens through real tamper/message/institutional acts with derived - signatures, never asserted psychic damage. Tamper leaves a persistent trace. + signatures, never asserted psychic damage. A negative self-trust consequence + settles only when the exact target later encounters the contradiction through + its ordinary read boundary without sufficient corroboration; tamper or plot + completion alone changes no psyche state. - Blowback is mandatory: every erode-self-trust plot ships a corroboration ending where the target is believed — self-trust regrows and the contradiction becomes ordinary observer-local evidence against the acting @@ -164,11 +171,13 @@ Owners: [aggregate-observer.md](../../mechanics/aggregate-observer.md) and - A generic GASLIGHT verb or any freely-pumped psyche dial. That is the placation ray the 2026-07-08 no-influence-token decision already refused; self-trust moves only through authored causal stories. -- Ending metadata that asserts destabilization without world acts — the - vending-machine failure this spec exists to prevent. -- Reusing tamper as a general evidence scrub. SuppressLogs keeps its narrow - detection-facing role; tamper targets one person's memory of their own - actions. +- Ending metadata that asserts destabilization without world acts, or applies + it before the target experiences the contradiction — the vending-machine + failure this spec exists to prevent. +- Reusing tamper as a general evidence scrub or letting it mutate canonical + history. SuppressLogs keeps its narrow detection-facing role; tamper targets + one person's memory of their own actions through a separate alterable + artifact. ### DEFERRED diff --git a/wiki/mechanics/plots.md b/wiki/mechanics/plots.md index 21ca934b..ac98cb9a 100644 --- a/wiki/mechanics/plots.md +++ b/wiki/mechanics/plots.md @@ -116,18 +116,26 @@ agents and the community can contribute libraries of them. lying to *unmake* someone: a story whose payoff is that the target stops trusting their own memory. Three vocabulary pieces carry it, and nothing else changes: - - A **`tamper` world act**: rewrite or erase one persisted record the bound - target authored or witnessed — a message they sent, a ticket they filed, - a change they pushed, a fix they made — so the surviving evidence - contradicts their memory. The act derives its signature from the record's - channel like every other act, and it leaves a persistent tamper trace in - the world. Tamper is aimed at one person's memory of their own actions; - it is not a general evidence scrub (the narrow SuppressLogs asset action - owned by social.md keeps that separate, detection-facing role). - - A **`self_trust` ending consequence**: endings may move the target's - self-trust axis (owned by social.md) alongside disposition and - obligation, negative in erosion endings and positive in corroboration - endings. Bounds are validated like every consequence. + - A **`tamper` world act**: alter or suppress one world-facing artifact + the bound target authored or witnessed — a message archive entry, ticket + view, change record, or filing revision — so the artifact they later + consult contradicts their memory. The original event, original record + payload, and any acquired observer evidence remain immutable simulation + truth. Tamper authors separate persisted custody binding the exact + original record, altered artifact or tombstone, acting persona, act tick, + carrier and channel, derived signature, and discovery state. It is aimed + at one person's memory of their own actions; it is not a general evidence + scrub (the narrow SuppressLogs asset action owned by social.md keeps that + separate, detection-facing role). + - A **`self_trust` ending consequence**: an ending may declare a bounded + delta alongside disposition and obligation, but erosion does not settle + merely because tamper fired or the plot reached its final beat. The exact + target must later encounter the altered artifact through its ordinary + carrier/read boundary, remember the contradictory event, lack sufficient + corroboration, and attribute the mismatch to their own memory. Only that + encounter applies the negative delta. Corroboration endings apply their + positive delta when another observer or the discovered tamper custody + backs the target's account. - The **`erode-self-trust` category** names the manipulation shape. The Marcus bar applies unchanged: a gaslighting plot succeeds through real tamper, message, and institutional acts the world can in principle notice @@ -139,7 +147,11 @@ agents and the community can contribute libraries of them. persona (personas.md contradiction records) plus suspicion. This is the deliberate symmetry: the same contradiction machinery that breaks your masks is what you are weaponizing against a person's self-model, and when - the story fails it snaps back into that ordinary machinery. + the story fails it snaps back into that ordinary machinery. Low self-trust + changes what the person does with knowledge — hesitation, confirmation + seeking, delayed or withheld reporting, and less reliable assigned work. + It never changes what physically happened or deletes exact acquired + evidence. - **A plot matches characteristics, not cast identity.** Schema 2 definitions declare one or more closed person roles and leverage kinds, plus only the access or switch capability gates their mechanism truly assumes. A matching @@ -298,17 +310,21 @@ failure/blowback). House rules: derive signatures from their channel, amount, event kind, and impact. - **Consequences are typed data only.** A plot cannot invent a new effect kind. Endings compose disposition, obligation, self-trust, and - leverage-serviced state; - messages, money, institutional history, and signatures exist only as the - results of world acts. If a story needs another effect, that is a spec - amendment, not a plot file. + leverage-serviced state; messages, money, institutional history, and + signatures exist only as the results of world acts. A negative self-trust + consequence is additionally conditional on the target's later exact + encounter with the contradictory artifact; final-beat metadata alone does + not apply it. If a story needs another effect, that is a spec amendment, + not a plot file. - **Erosion earns its cruelty.** An erode-self-trust plot must reach its - erosion through tamper acts against records the target really authored or - witnessed, and it must ship the mandatory corroboration blowback ending. - A story that only sends unsettling messages has not manufactured the - memory-versus-evidence gap that makes gaslighting gaslighting, and fails - review; a story with no path where the target is believed fails - validation. + erosion through tamper acts against artifacts for events the target really + authored or witnessed, followed by an ordinary target-facing read or + confrontation where corroboration is absent. It must also ship the + mandatory corroboration blowback ending. A story that only sends unsettling + messages, mutates canonical history, or ends before the target encounters + the contradiction has not manufactured the memory-versus-artifact gap that + makes gaslighting gaslighting, and fails validation; a story with no path + where the target is believed fails validation. - **A debt payment reaches the debt.** A plot that says a payment services Debt leverage sends that settlement to the target-relative creditor and retires that target's modeled debt flow. Paying an unrelated external @@ -416,14 +432,17 @@ the plot remains held. traverses the same slot reservation, Thought reservoir, resource, world-act, signature, blocking, and failure paths as a manual submission; no policy executes a generic leverage-servicing shortcut. -12. (ADOPTED 2026-07-27, pending) The `tamper` world act rewrites or erases - one persisted target-authored or target-witnessed record, derives its - signature from that record's channel, and leaves a persistent tamper - trace discoverable by audit, a corroborating observer, or intel - processing. Endings may carry a bounded `self_trust` delta applied by the - shared executor and round-tripping through save. At least one - erode-self-trust plot ships; its validation rejects a definition without - tamper acts or without a corroboration ending, and its blowback path - restores self-trust, creates observer-local evidence against the acting - persona, and raises suspicion. There is no generic gaslight verb: the - category is reachable only through authored plots. +12. (ADOPTED 2026-07-27, pending) The `tamper` world act changes or + suppresses one target-authored or target-witnessed world-facing artifact + without rewriting the canonical event, original payload, or acquired + evidence. It persists exact original/altered custody, acting persona, act + tick, carrier/channel, derived signature, and discovery state. Endings may + carry a bounded `self_trust` delta that round-trips through save, but a + negative delta settles only after the exact target encounters the altered + artifact through its ordinary read boundary without sufficient + corroboration. At least one erode-self-trust plot ships; validation rejects + a definition without tamper, without that target-facing encounter, or + without a corroboration ending. Its blowback path restores self-trust, + creates observer-local evidence against the acting persona, and raises + suspicion. There is no generic gaslight verb: the category is reachable + only through authored plots. diff --git a/wiki/mechanics/social.md b/wiki/mechanics/social.md index a9afcf5e..1acf0172 100644 --- a/wiki/mechanics/social.md +++ b/wiki/mechanics/social.md @@ -86,18 +86,22 @@ taught, contradiction by manufactured contradiction, that their own memory cannot be trusted. Rules: - Self-trust starts steady and stays steady by default. Nothing generic - moves it: no verb, no token, no dial. It changes only as the typed - `self_trust` consequence of an authored plot whose beats performed real - tamper and message acts against records the person authored or witnessed - (plots.md owns the act and consequence contract). This is the same - discipline as the 2026-07-08 no-influence-token decision — self-trust is - not trust toward you, and there is no placation-ray equivalent for it. + moves it: no verb, no token, no dial. A negative typed `self_trust` + consequence belongs only to an authored plot that changed a separate + world-facing artifact for an event the person really authored or witnessed, + then let that person encounter the contradiction through its ordinary + carrier/read boundary without sufficient corroboration (plots.md owns the + act and encounter contract). Tamper or plot completion alone changes no + self-trust. This is the same discipline as the 2026-07-08 + no-influence-token decision — self-trust is not trust toward you, and there + is no placation-ray equivalent for it. - **A shaken person is a worse witness.** What they see with their own eyes converts attention into suspicion at reduced weight, and they hesitate to file or report what they cannot corroborate — the gaslit custodian who watched your robot cross the hall tells no one, because last month the - logs proved the thing they were certain of never happened. This is the - strategic payoff and it is bounded [TUNE]. + logs contradicted the thing they were certain happened. This changes what + they do with exact knowledge; it never rewrites the event or removes their + acquired evidence. This is the strategic payoff and it is bounded [TUNE]. - **A shaken person is also a worse tool.** Erosion cuts both ways: a destabilized person re-checks, second-guesses, and mis-executes, so their asset task reliability degrades below their knowledge-level baseline @@ -280,13 +284,17 @@ retirement, burning, and reopening are bound to the separate PERSONAS view. ### Adopted 2026-07-27, pending implementation: self-trust -9. Every person carries a serialized self-trust value that starts steady, - changes only through the typed plot consequence, and round-trips through - save. A shaken person's own witnessed events convert to suspicion at - reduced weight and their asset task reliability sits below their - knowledge-level baseline; corroboration or tamper-trace discovery - restores self-trust and produces the ordinary evidence and suspicion - blowback. No generic action mutates the axis. +9. Every person carries a serialized self-trust value that starts steady + and round-trips through save. A negative typed plot consequence settles + only after that exact person encounters a contradictory world-facing + artifact through its ordinary read boundary without sufficient + corroboration; tamper and plot completion alone do not move the axis. A + shaken person's own witnessed events convert to suspicion at reduced + weight and their asset task reliability sits below their knowledge-level + baseline, while the original event and acquired evidence remain exact. + Corroboration or tamper-custody discovery restores self-trust and produces + the ordinary evidence and suspicion blowback. No generic action mutates + the axis. ### Operations interface receipts