Retrofit the evidentiality pamphlet against current canon

[dcellison]

Summary

• Retrofit all eleven chapters of Evidentiality in Phi against the current manual, primer, particle entries, modifier-first syntax, and completed base vocabulary.
• Restore the four particles to one bounded job each: hi claims direct perception, ke inference from case-specific evidence, ti a received report, and ho background assumption.
• Separate source marking from truth, reliability, confidence, honesty, politeness, and moral character throughout explanations, drills, exercises, and the quick reference.
• Replace the prescribed "honest journal" with an optional journal of sources and rename its source file accordingly.
• Synchronize the pamphlet catalogue, the direct three-slots dependency, the manual's journal pointer, and every SEM-09D handoff and roadmap resume point.

Language corrections

• Remove the ranking of hi as strongest and ho as humblest.
• Correct the unmarked clause as an assertion that names no source, not a hidden direct evidential, confidence claim, or fifth source choice.
• Remove future hi for promised presence; an event that has not happened cannot yet be witnessed.
• Keep past hi as a claim about the original perception without treating it as a guarantee of present memory.
• Keep another speaker's hi from transferring through a report while allowing a listener's later direct perception to establish a new hi clause.
• Put matrix ti after the complete mena ... meno complement and before its matrix predicate.
• Keep ho apart from confidence and modal po, and keep questions apart from any invented courtesy hierarchy.
• Repair the journal's missing past tense, the exercise that called valid omission an underclaim, the stale rain predicate in the three-slots answer key, and several English translations that overstated the Phi.

Editorial pass

The Humanizer and voice pass removed the recurring moral bookkeeping metaphors around "debt," "coin," and "earned" source marks. It also reduced claim as a batch workhorse from 54 uses to one while preserving the canonical term "claimed source" where it belongs. The stamping scan now finds only deliberate repetition: the fixed Slot 1 sequence, the four source definitions, the snake-at-the-well motif, and exercise-answer echoes. New prose contains no em or en dashes, curly quotes, banned inflation vocabulary, copula-avoidance phrases, inline-header lists, or -bearing constructions.

Vocabulary and tracking

• No root or lexical decision was needed.
• The general pamphlet uses no optional module term.
• Lexicon counts remain 1,275 entries, 261 optional roots, and 393 module memberships.
• The phonetic-neighbour set remains the same 250 pairs; the generated baseline changes only because revised examples alter corpus-attestation counts and ordering.
• SEM-09D now resumes with pamphlets/three_slots/, followed by the other seven pamphlets and the current-document sweep.

Checks

• python3 scripts/validate_examples.py (1,275 entries, 0 errors, 0 warnings)
• python3 scripts/test_vocabulary_schema.py (22 tests)
• python3 scripts/content_vocabulary_decisions.py --check (63 batches, 235 decisions)
• python3 scripts/test_content_vocabulary_decisions.py (6 tests)
• python3 scripts/test_name_forms.py (20 tests)
• python3 scripts/content_catalogues.py --check
• python3 scripts/audit_phonetic_neighbors.py --output /tmp/phonetic_neighbors_baseline_final.txt plus a clean baseline diff
• python3 scripts/generate_reference.py with no generated-reference drift
• python3 scripts/build_site.py (98 pamphlet pages; full site build succeeded)
• git diff --cached --check