Verifying oral knowledge means comparing carriers who learned it independently and settling disagreements by a reference, not by a headcount. If three carriers each get an item wrong with probability 0.1 and their errors are independent and identical, a majority vote is wrong with probability about 0.028; but if two of them learned from the same teacher they count as one witness and the majority is wrong with probability 0.1, no better than a single carrier.
Part of Recording & Backup Systems
Verification and version control are the procedures that establish which wording of a preserved item is correct, record where each version came from, and retire superseded ones.
Majority Voting Helps Only Against Independent Errors
A common rule is “if most carriers agree, that version is right”. It is worth working out when that is true. The numbers below are illustrative assumptions, not measurements: p is the probability that one carrier has a given element wrong, and the worst case is assumed in which every wrong carrier holds the same wrong value, so that wrong carriers agree with each other. If wrong carriers hold different wrong values, they split the vote and the majority does better than shown.
Three independent carriers, majority of two or more wrong:
P(majority wrong) = 3p²(1 − p) + p³
| p (one carrier wrong) | 3p²(1 − p) | p³ | Majority wrong |
|---|---|---|---|
| 0.05 | 3 × 0.0025 × 0.95 = 0.007125 | 0.000125 | 0.00725 |
| 0.1 | 3 × 0.01 × 0.9 = 0.027 | 0.001 | 0.028 |
| 0.3 | 3 × 0.09 × 0.7 = 0.189 | 0.027 | 0.216 |
| 0.5 | 3 × 0.25 × 0.5 = 0.375 | 0.125 | 0.5 |
At p = 0.1 the vote cuts the error from 0.1 to 0.028, a factor of 0.1 ÷ 0.028 = 3.6. At p = 0.5 it cuts nothing: when each carrier is a coin flip, a vote is a coin flip. With five independent carriers at p = 0.1, the majority (three or more wrong) fails with probability 10 × 0.1³ × 0.9² + 5 × 0.1⁴ × 0.9 + 0.1⁵ = 0.0081 + 0.00045 + 0.00001 = 0.00856.
Shared Teachers Collapse The Gain
Now let carriers A and B both have learned the item from teacher T, who has the element wrong with probability 0.1, while C learned independently and is wrong with probability 0.1. If T is wrong, A and B inherit the error and are both wrong, so the majority is wrong whatever C says: P(majority wrong) = 0.1, the same as a single carrier. Counting heads gave 0.028 when the three were independent and 0.1 when two shared a teacher.
The practical rule is to weight by independent lineages, not by headcount. Textual critics weigh manuscripts the same way, treating copies from one ancestor as one witness (the method is associated with Lachmann, and is recalled here, not re-checked). So record who learned from whom in the register (Redundancy: Multiple Carriers).
Estimating p In Your Own Community
Run a cross-recitation: choose a set of independent elements from the reference, have each carrier recite them separately, and count mismatches against the reference. If 4 of 40 elements differ, the estimate is p = 4 ÷ 40 = 0.10. Its uncertainty is large: the standard error is √(0.1 × 0.9 ÷ 40) = √0.00225 = 0.047, so a plausible range is roughly 0.05 to 0.15. Forty elements will not distinguish 0.05 from 0.15 reliably. Repeat the test before relying on any particular p.
Method: Settle A Disagreement
Step 1: Classify the disagreement.
| Case | Treatment |
|---|---|
| All agree | Check lineage. If they share a teacher, agreement is one witness. Check against the reference anyway if one exists |
| Minority differs | Ask what the dissenter is remembering. A rare context may be the right one |
| Tie | Reference, physical test or an independent third carrier |
| No reference, split carriers | Mark disputed, keep both readings, use the safer course for safety items until resolved |
Step 2: Use the evidence ladder. From strongest to weakest: a physical or practical test; a verified written reference; carriers confirmed independent; one carrier.
Step 3: Record the version. Give each fixed item an identifier, a date, the names of those who verified it, its parent version and the reason for any change. Withdraw superseded versions explicitly.
Step 4: Add internal checks. Counts (“five steps”), the list of rhyme words, and the first and last lines catch omission and reordering. They do not catch substitution of one number for another (Rhythm, Rhyme & Song).
Step 5: Update every carrier after a ruling, and log it.
Worked Dispute: The Carrier Article’s Own Example
The carrier article says that if two carriers say one minute and one says three, “1 minute is correct” because the item is a sea-level purification. That decision rests on the context (“sea level”), not on the vote. The same article elsewhere gives three minutes above 2000 m, so the dissenter may be remembering the altitude case, which is correct there. A vote would have deleted a correct variant. The right procedure asks the dissenter where they learned it, and then checks the figure against Water Purification.
The same article also says the checker’s version is the standard and, separately, that the majority rules. These conflict whenever the checker is in the minority. Use this order instead: reference first, tested practice second, independent carriers third, the checker’s memory fourth (Roles: Teller, Listener, Checker).
Doses and safety quantities are never settled by a vote
A majority can agree on a wrong dose or limit. Do not decide any medicine dose or poison threshold by carrier vote or by who is more confident. Mark it disputed, take the safer course while it is checked against a written or tested source, and do not alter a dose on the basis of recited memory.
Common Mistakes
- Counting heads without lineage — shared teachers make agreement look stronger than it is.
- Deleting the dissent — keep minority readings in the log until resolved.
- Trusting agreement as proof of truth — agreement shows consistency, not correctness; only a reference or a test shows correctness.
- Recency as authority — the newest verified version is not necessarily the right one.
What’s Next
- Learn next: Correction Protocols — the in-session version of this procedure
- Combine with: Decision Archives — recording rulings and their reasons
Unverified references
The textual-criticism principle attributed to Lachmann is from recollection. The carrier probabilities p are assumptions used to show the arithmetic, not measured rates; the arithmetic itself is shown so it can be checked, and assumes identical wrong values among the wrong carriers.