EV_MULTI_007 — Reported buyer evaluation of multilingual quote batch processing

Structured call-learning evidence supplied by Gabriel Paunescu, CEO, reporting buyer interviews about multilingual quote processing in production. No live database, customer system, raw quote record, transcript, recording, screenshot, or export was inspected during this authoring task.

kind: evidence_catalog
schema_version: 1
evidence:
  - id: EV_MULTI_007
    type: call_learning
    title: 'Reported buyer evaluation of multilingual quote batch processing'
    source: >-
      Interactive Codex create-proof task for P_00014 and S_00006 on
      2026-08-01. Gabriel Paunescu reported buyer interviews covering more than
      25 production customers. No separate stable interview, transcript,
      recording, or database-export reference was supplied.
    publisher_or_owner: 'Gabriel Paunescu (CEO), reporting buyer interviews'
    published_at: null
    accessed_at: 2026-08-01
    supports:
      - >-
        Gabriel Paunescu reported that buyers used the multilingual quote
        upload, language-normalization, extraction, and batch-intake workflow
        in production across more than 25 customers collectively spanning
        M1_V1, M1_V2, M1_V3, M1_V4, M1_V5, M1_V6, and M1_V7.
      - >-
        The reported evaluation used one batch of 1,000 production quote
        records and real database values as the comparison truth set. A quote
        passes only when every required field matches that truth set; the
        buyers accepted no more than 5 percent, or 50 of 1,000, mismatched
        complete quotes.
      - >-
        The contributor reported that start and end timestamps covered the
        1,000 quotes and that buyers described the quotes as being sent three
        times faster than normal.
    scope: >-
      Reported production use across more than 25 customers collectively
      spanning M1_V1, M1_V2, M1_V3, M1_V4, M1_V5, M1_V6, and M1_V7; one
      1,000-quote multilingual batch evaluated against real database values.
    confidence: medium
    notes: >-
      User-provided wording preserved faithfully includes: "all verticals.
      this is a general function. over 25 customers in production"; "buyer
      during buyer_interview"; "yes. it's a great result; per quote; real data
      in db; 1000 quote sample set; forget about it"; "correct. it's
      accetable"; "i Gabriel Paunescu is reporting the buyer interviews";
      "no. they get sent 3x faster than normal"; and "start/end timestamps. It
      covered 1000 quotes. batch size = 1000." Because Gabriel Paunescu is
      reporting buyer interviews rather than answering as the buyer, this is
      call-learning evidence. The exact interview dates, calendar observation
      period, account identities, language list, required-field list, truth-set
      construction procedure, mismatch adjudication procedure, exact prior and
      observed durations, aggregation method behind "3x faster," and treatment
      of partial batch failures were not supplied. The raw database values and
      timestamps were not inspected. The 5 percent value is a buyer-accepted
      maximum per complete quote, not an independently verified observed error
      rate. This evidence does not establish universal language coverage,
      automatic correctness, implementation details, or capacity beyond the
      reported 1,000-quote batch.