Hubris Self Thinking V1

Hubris Self Thinking 2.14.0 — private draft Reusable NPC continuity guided by your character cards: individual motives, boundaries, relationships and available abilities. It encourages direct dialogue, subtext, disagreement and changes grounded in experience. Natural dialogue and long-term emotional growth still require live review. Seven verified rounds since 2.7.28: 1. Expressive reactions. The affect gate no longer demands a redundant emotion word beside a face verb, so sobbed, scowled, blushed, wept and dozens of ordinary reactions register again. Narration vocabulary grew by 34 rules and spoken feelings by 18. Every honesty veto (hedge, contrast, sarcasm, negation, pretence) is unchanged. 2. Universal emoji compatibility. Any emoji that can stand for a person's reaction now counts, in every skin tone and gender variant — about 715 forms instead of 114 — and gestures (shrug, facepalm, wave, applause, handshake, folded hands, raised hand, blown kiss) can be cast by the script. Objects, places, flags and family sequences stay recognised and preserved where they are, but never become a character's face. 3. Prose-derived personality. The personality system previously read only @tag lines from a card's description, so cards written as ordinary prose contributed nothing character-specific and every NPC fell back on the model's default voice. Anchors (personality, voice, flaw, motive, attention, decision, softspot) are now derived from the card's own sentences when a tag is silent. Every derived value is a literal span of the author's text — never a paraphrase or invention. An authored @tag always wins. 4. Review hardening. Five bugs in the new code, each reproduced before the fix: "but she wasn't happy" still received a happy face; "sick of this" read as nausea; clapping a hand over a mouth read as applause; raising a hand to strike read as a raised hand; gagging someone read as nausea. 5. NPC depth, scenario-neutral. Supportive and hostile action vocabulary widened so an act written in natural prose registers instead of needing an exact keyword. Harm a character suffers from a third party now raises their own distress, so a knife at their throat in the narration reaches the model; it still grants no relationship change. Where a card never describes how someone behaves under pressure or toward a trusted person, the tracked state itself is stated instead of nothing, so the arc is visible. 6. Earned trust. Relationship changes are owned by the accepted-receipt system and no keyword can grant them, which is correct — but the recognizer feeding that system was four verbs ("help/save/protect/heal"), verb-initial, with the character's name immediately after. So protecting someone earned trust while shielding, carrying or sheltering them earned nothing. The recognizer now covers about thirty genre-neutral physical acts. Every safety check is unchanged: speech, and negated, attempted or bystander-directed acts, still earn nothing. Three distinct kindnesses now move a character from reserved to openly trusting. 7. Cache-safe context. Under Optimized Context the platform discards a context-modifier addition unless it is a pure append. This script inserted its detailed packet mid-context, so on those models it could be dropped silently. When that regime is detected the packet is now appended; classic placement is unchanged. The compact packet always travelled via the Author's Note and was never at risk. /hst status now reports which regime an adventure is running. Rounds 5-7 were verified across high fantasy, science fiction, modern thriller and historical settings. Emojis are on by default and can be changed separately in each adventure: /hst emojis off — disable emojis in future replies. /hst emojis on — enable them again. /hst dialogue natural — select the natural dialogue register. /hst inspect <full name> — inspect one character's tracked state. /hst status — settings, plus the active context regime. /hst help — list commands. /hst off and /hst on — disable or enable the script while retaining its records. For exact dialogue, use Say mode or straight double quotation marks around speech in Do mode. In the platform tests, single-quoted Do speech could be changed before the script received it. Verification: 55 behaviour, 17 regression, 8 personality, 13 depth and cross-genre, 13 earned-trust and 3 context-delivery checks all pass on this build, plus a 5,216-form emoji coverage audit. Failures were reproduced on the previous build before each fix. Worst hook time 12 ms against the 2,000 ms budget. Whether the resulting dialogue reads as genuine over long play is a narrative judgement these checks cannot settle and still requires live review. This private draft has not been published; existing added-script attachments do not automatically receive it. Test copies run the verified new code.