================================================================
 TETRACTA — REPRODUCIBILITY MANIFEST (public evidence bundle)
================================================================
 Generated: 2026-06-13
 Purpose:   Pin each public evidence artifact to a checksum so anyone
            can verify the file they downloaded is the file we published.
 Policy:    These artifacts contain measured RESULTS only — abundantly.
            What is withheld is only the IP: the formula / sink design,
            kernel internals, exact learning rates, schedule, SPC
            threshold calibration, tokenizer internals, and quant
            performance figures. "What" is shared in full; "how" is not.
----------------------------------------------------------------
 Verify on your machine:
   Linux/macOS:   md5sum -c tetracta-reproducibility-manifest.txt
   Windows (PS):  Get-FileHash <file> -Algorithm MD5
 Note: this manifest cannot hash itself, and server-config files (e.g.
 .htaccess) are intentionally outside the checksummed bundle below.
----------------------------------------------------------------

# md5 (lowercase hex)  *  filename
714233149b17f9396813ceab46028cb0  tetracta-results-report.md
89b31fbf491a41a14ef247b0b5801b06  tetracta-experiments.csv
c213e80e497ad8386ecc049ed15b7a4e  tetracta-convergence-3b.csv
90bb4475ed86d277f4d09b8612dc6a5c  tetracta-stability-pairs.csv
36bbe124c661802200f6056c6ce86099  tetracta-kernel-equivalence.csv
5329156fd2c560c23a062d309ea6661f  tetracta-scale-curve.md
01538b1da58427729c792794a1a1efc1  tetracta-scale-curve.csv
7d22168910979c354d4ef64401385ffe  tetracta-stability-benchmark.md
47364cd5174bb6ba78fc3ae57e4cebb9  tetracta-stability-benchmark.csv
36ffeccda06297673830f322d53874a5  tetracta-methodology-onepager.md
84a4c16289554a2935410aa0ecdd8cb3  tetracta-quant-methodology-card.md

----------------------------------------------------------------
 ABOUT THE INTERNAL EVIDENCE CHAIN (not in this public bundle)
----------------------------------------------------------------
 Behind these summaries sits a sealed internal chain: raw per-step
 training logs (JSONL), deterministic extractor scripts, and a full
 md5 manifest of every run. Headline numbers regenerate bit-for-bit
 from the raw logs; a deliberately-corrupted control is included and
 correctly fails the equivalence test. Under NDA, qualified investors
 and partners can inspect this chain in full.

 The training-data token stream used for all A/B runs is itself pinned
 by checksum in our internal records (single shared corpus = fair A/B).

 © 2026 Tetracta — Rational Attention(TM), method patent application in preparation.
================================================================