Research · Four-arm workflow · Diagram C3 · Appendix A5

Arm C3: a note-taker that never steers

Every C2 decision box is printed word for word and stays plain. Only RECORD changes: a consistency note, for the record only. The crossed-out reverse is the finding. Nothing it writes ever reaches a decision.

Arm C3: a note-taker that never steers Arm C3 keeps every C2 decision box unchanged and writes a consistency note to the Evidence Trace for the record only, with no path back into any decision. while an answer could still change advice and the question limit allows another question. 1 HEAR Model only turns Thai into facts. It decides nothing. 2 BUILD THE LIST Sort signs: support / contradict / not yet asked. Band: Likely, Possible, Cannot exclude. No number. 3 CHOOSE WHAT TO ASK Rank by whether the answer would change what the farm does. also estimate the expected split; also reuse saved first-turn. 4 ASK One question per message. Questions not asked are written with why. 5 DECIDE WHETHER TO STOP Stop when no answer could change advice. Question limit: 4 + 1 hard stop 10. 6 DECIDE WHAT MAY BE SAID Name one disease, an honest tie, or send to the veterinarian. Never a percentage. 7 SAY Model writes Thai from approved wording. Second and last time. 8 RECORD FOR THE RECORD ONLY Evidence Trace, plus a consistency note. One arrow in, no arrow back to any decision. 9 REFUSAL / VET DOOR BY COVERAGE no arrow back to any decision X METAPHOR NOTE RECORD X DEC it writes, it never steers RMC GRID 48x3=144 · C3 acc 68/144 beside B 74/144 · RMC GRID 48x3=144 · C3 wrong reportable flag 37/144 beside B 46/144 · answer key follows Arm B LEGEND Stage Record Main flow Loop / write

Full sentences · matched to the numbered chips

  1. 1

    HEAR

    The model is used only to turn the Thai into facts. It decides nothing.

  2. 2

    BUILD THE LIST

    For each disease in the approved catalogue, sort its listed signs into supporting / contradicting / not yet asked. Then give it a plain word from a fixed 8-row table: Likely, Possible, or Cannot exclude. Never a number. Order the diseases still on the list. Reportables stay in the protected set and are always checked.

  3. 3

    CHOOSE WHAT TO ASK

    Rank every possible question by whether its answer would change what the farm does. If two diseases lead to the same Farm Operator actions, do not spend a question splitting them. Pick one.

    Also estimate the expected split from possible answers. Also reuse the saved first-turn work instead of doing the work again.

  4. 4

    ASK

    One question per message. Every question not asked is written down with its reason.

  5. 5

    DECIDE WHETHER TO STOP

    Stop when no remaining answer could change what we advise the farm to do. Also stop when the question limit is reached: 4 answers, plus 1 extra safety question. So about five questions at most in an ordinary case. There is a hard stop at 10 asking turns.

  6. 6

    DECIDE WHAT MAY BE SAID

    Name one disease, state an honest tie, or send to the veterinarian. Never show a percentage.

  7. 7

    SAY

    The model is used a second and last time, only to write the Thai from approved wording.

  8. 8

    RECORD

    Evidence Trace (the full written record), plus a consistency note. For the record only. One arrow in, no arrow back to any decision.

  9. 9

    REFUSAL / VET DOOR

    REFUSE BY COVERAGE. Does the approved catalogue cover this Health Event at all? If not, send to the veterinarian. No confidence number is needed.

Caption

Paperwork, not the machine

C3 changed the paperwork, not the machine. Identical rows are the proof it never steered. RMC GRID 48x3=144: C3 acc 68/144 beside B 74/144; C3 wrong reportable flag 37/144 beside B 46/144; C3 mean questions 3.87 beside B 3.58. The same grid prints C3 identical to C2 on those three numbers. That is because nothing it writes reaches a decision. The answer key follows Arm B.

Changed from

RECORD only

Changed from C2: C3 keeps all of C2. Every decision box stays put. It adds only a consistency note in RECORD, for the record only. The write is one way. There is no path back. It writes. It never steers.

How to read

Marks on this drawing

  • Dashed RECORD: Evidence Trace plus a consistency note, for the record only
  • Crossed stub: no arrow back to any decision
  • Grey strip: RMC GRID 48x3=144 · C3 acc 68/144 beside B 74/144 · answer key follows Arm B

Research only - nothing is deployed - no percentage ever reaches the Farm Operator - the 14 Disease Profiles are vet-reviewed - reviewed, not signed.