feat(jepa_t_ingest): Wave-14a L-S50 — plaintext → ternary triplet pipeline for JEPA-T

[gHashTag]

Closes #807

Summary

Adds crates/jepa_t_ingest/ — a Rust-only (R1 CROWN) crate that streams plaintext corpora into ternary-quantized triplet sequences for JEPA-T training on Trinity silicon (Wave-14a L-S50).

Quantizer — Wave-9b RTL byte-for-byte match

pub fn quantize_phi_prior(fp_q15: i16) -> i8

Threshold: φ⁻² in Q1.15 = 12533 (0x30F4):

• if fp_q15 >= +12533 → +1
• if fp_q15 <= -12533 → -1
• else → 0

Deliverables

File	Description
Cargo.toml	edition 2021, Apache-2.0, author Dmitrii Vasilev
src/lib.rs	quantize_phi_prior, IngestConfig, Triplet, ingest_text
src/bin/jepa_t_ingest.rs	CLI: --input corpus.txt --output triplets.bin
tests/quantize.rs	Boundary tests ±12532→0, ±12533→±1, 0, ±0x7FFF; exhaustive 65536-input scan
tests/ingest.rs	Golden corpus byte-compare (3 triplets, 192 bytes each)
README.md	Full API docs + quantizer table

cargo test -p jepa_t_ingest output

running 8 tests (unit)
test unit_tests::ingest_empty_returns_empty ... ok
test unit_tests::ingest_produces_valid_ternary ... ok
test unit_tests::ingest_short_returns_empty ... ok
test unit_tests::quantize_max_values ... ok
test unit_tests::quantize_negative_boundary ... ok
test unit_tests::quantize_positive_boundary ... ok
test unit_tests::quantize_zero ... ok
test unit_tests::triplet_to_bytes_length ... ok
test result: ok. 8 passed; 0 failed

running 10 tests (tests/ingest.rs)
test empty_corpus_returns_empty ... ok
test all_values_are_ternary ... ok
test golden_corpus_triplet_count ... ok
test golden_first_triplet_anchor ... ok
test golden_first_triplet_negative ... ok
test golden_first_triplet_positive ... ok
test golden_total_bytes ... ok
test large_stride_still_produces_triplets ... ok
test single_token_returns_empty ... ok
test triplet_serialises_to_192_bytes ... ok
test result: ok. 10 passed; 0 failed

running 12 tests (tests/quantize.rs)
test boundary_minus_12533_is_negative_one ... ok
test boundary_minus_12532_is_zero ... ok
test boundary_plus_12532_is_zero ... ok
test boundary_plus_12533_is_positive_one ... ok
test max_i16_is_positive_one ... ok
test min_i16_is_negative_one ... ok
test mid_range_negative_is_zero ... ok
test mid_range_positive_is_zero ... ok
test one_above_threshold_is_positive_one ... ok
test one_below_negative_threshold_is_negative_one ... ok
test output_always_ternary_for_all_i16 ... ok
test zero_is_zero ... ok
test result: ok. 12 passed; 0 failed

Doc-tests jepa_t_ingest: 2 passed; 0 failed

TOTAL: 32 tests, 0 failures

Quantizer Parity Confirmation

Input	Expected	Actual	Status
+12532	0	0	✓ PASS
+12533	+1	+1	✓ PASS
-12532	0	0	✓ PASS
-12533	-1	-1	✓ PASS
0	0	0	✓ PASS
+0x7FFF	+1	+1	✓ PASS
-0x8000	-1	-1	✓ PASS
all 65536 i16 inputs	ternary	ternary	✓ PASS (exhaustive)

R1 CROWN

No Python files in crates/jepa_t_ingest/. Rust ONLY.

References

• Wave-9b RTL phi_prior_quantizer.v
• DOI: 10.5281/zenodo.19227877
• φ² + φ⁻² = 3; ternary alphabet {−1, 0, +1}