Debate Methodology as a Validation Protocol
Chapter Sixty-Two
Syllabus topic Module 2, "Debate methodology and validation", "Inference engines"
Pages 213 to 216 of 378
In one line
Nyāya's debate rules are a validation protocol: they say what may be claimed, what counts as support, what moves are allowed, what counts as a violation, and how the exchange ends.
In the wording you can write in an examination: the Nyāya apparatus for disputation constitutes a validation protocol, that is, a procedure by which a claim is tested by an adversary under stated rules. Its components are a required form for the claim and its support, a fixed set of admissible means of knowledge, an enumerated set of defects in a reason, an enumerated set of improper moves, and enumerated grounds on which a party is declared defeated.
The five parts of a protocol
Any procedure for settling disagreements must have these five, and Nyāya has all five explicitly.
| Part | What it answers | Nyāya's version |
|---|---|---|
| The form of a claim | how must a position be stated? | the five members, 1.1.32 |
| Admissible support | what counts as evidence? | the four pramāṇas, 1.1.3 |
| Detectable defects | what makes support bad? | the five hetvābhāsa, 1.2.4 |
| Improper moves | what may a participant not do? | chala and jāti, 1.2.10 and 1.2.18 |
| Termination | how does it end? | the nigrahasthāna of Book V.2 |
Missing any one and the procedure fails in a specific way. Without a form, a party can shift ground. Without admissible support, every dispute becomes a dispute about what counts as a reason. Without defects, a bad argument cannot be named as bad. Without improper moves, the exchange degenerates. Without termination, it never ends.
Why adversarial testing is the design
The whole apparatus assumes an opponent whose job is to find the fault. That is a choice and it has consequences worth stating.
What it buys. A claim that survives a hostile examination has been tested in a way no amount of self-checking achieves. The arguer does not choose the attacks, so the attacks are not the ones the arguer has already thought about.
What it costs. It needs a second party, and the second party must be competent and must be playing. Vidyabhusana's note on the Buddhist opponent shows the first difficulty and the distinction between vāda and jalpa shows the second.
And it has no self-test. Nothing in the machinery examines itself. A claim nobody attacks passes, however bad it is. That is the standing weakness of any adversarial validation and it is the reason modern practice adds mechanical checks alongside human review.
Three modern protocols with the same five parts
Peer review of a paper
| Part | Its form |
|---|---|
| form of a claim | the paper, with its stated contributions and its method section |
| admissible support | experiment, proof, and citation of accepted results |
| detectable defects | the reviewer's checklist: unsupported claim, flawed experiment, missing baseline |
| improper moves | reviewing the author rather than the work; demanding a different paper |
| termination | the editor's decision |
Debate Methodology as a Validation Protocol
Where it is weaker than Nyāya. The list of defects is not enumerated anywhere, so two reviewers apply different lists and neither can be told they have applied none.
A test suite
| Part | Its form |
|---|---|
| form of a claim | the assertion, which names the expected value |
| admissible support | the actual run of the code |
| detectable defects | a failing assertion, and a test that passes when the code is broken |
| improper moves | asserting on incidental output, or testing the mock rather than the code |
| termination | the suite goes green, or it does not |
Where it is stronger than Nyāya. It is mechanical, so it needs no competent opponent, and it runs on every change. That is the modern addition that the classical apparatus has no equivalent of, and it is why this book's own claims are checked by programs rather than argued about.
A code review
| Part | Its form |
|---|---|
| form of a claim | the pull request, with its description of what it changes and why |
| admissible support | the diff, the tests, and the measurements |
| detectable defects | the review checklist |
| improper moves | objecting on style against a claim about correctness, which is jāti exactly |
| termination | approval, or a request for changes |
Where it repeats Nyāya's weaknesses. It needs a competent reviewer who is actually engaging, and a review that finds nothing is indistinguishable from a review that did not look.
Worked example: a claim through the whole protocol
The claim. This change makes the endpoint twice as fast.
Form. Stated in five members. Proposition: this change makes the endpoint twice as fast. Reason: because it removes the per-request database lookup. Example: whatever removes a per-request database lookup on this endpoint halves its latency, as the change to the neighbouring endpoint last month. Application: so is this change: it removes that lookup. Conclusion: therefore it makes the endpoint twice as fast.
Admissible support. The measurement is pratyakṣa. The neighbouring endpoint's result is the familiar instance. The claim that the lookup was per-request rests on reading the code, which is pratyakṣa of the code rather than of the behaviour.
Defects, checked. Is the reason erratic: could removing that lookup fail to halve the latency? Yes, if the lookup was not the dominant cost. The connection is erratic and the argument fails, which the form exposed at the example member.
Repair. Narrow the mark: because it removes the per-request database lookup, which the profile shows accounted for half the latency. Now the connection is defensible, and the cost of the repair is a profile, which is the second observation [The Five Members Worked, Three Times] predicts.
Debate Methodology as a Validation Protocol
Improper moves to watch for. An objection that the variable names are poor is jāti: relevant to something, irrelevant to this claim.
Termination. The claim is established or the change is not merged.
What the protocol does NOT do
It does not find claims. Nothing in it tells you what to test. The five extra members of [The Five-Member Syllogism] include jijñāsā, inquiry, which is an attempt to bolt a discovery stage onto the front, and it is not part of the standard five.
It does not guarantee truth. A claim can survive every check and be false, because the vyāpti rests on observation and observation does not establish universals.
It does not work without an opponent. This is the one to say last, because it is the sharpest criticism and it is the one modern practice answers. A protocol whose only detector is a person can be satisfied by nobody looking. A machine check cannot be satisfied by nobody looking, which is why every claim in this book about a number, a table or an algorithm is produced by a program that runs.
Quick revision
- Five parts of a validation protocol: the form of a claim, admissible support, detectable defects, improper moves, and termination.
- Nyāya has all five explicitly: 1.1.32, 1.1.3, 1.2.4, 1.2.10 with 1.2.18, and Book V.2.
- Missing each one fails in a specific way: shifting ground, disputes about what counts as a reason, unnameable bad arguments, degeneration, and no end.
- Adversarial design buys attacks the arguer did not choose and costs a competent opponent who is playing.
- Peer review, a test suite and a code review all have the same five parts. The test suite is the one that needs no opponent.
- The protocol does not find claims, does not guarantee truth, and cannot detect a failure nobody looked for.
Test yourself
1. Name the five parts of a validation protocol and Nyāya's version of each.
The form of a claim, the five members. Admissible support, the four pramāṇas. Detectable defects, the five fallacies of the reason. Improper moves, quibble and futility. Termination, the occasions for rebuke.
2. What does a protocol lose if it has no termination condition?
It never ends. Without enumerated grounds of defeat there is no point at which one party has lost, so the exchange can be prolonged indefinitely by a participant who will not concede.
3. Give the greatest strength and the greatest weakness of adversarial validation.
Its strength is that the attacks are chosen by someone other than the arguer, so they are not the ones already anticipated. Its weakness is that it requires a competent opponent who is actually engaging, and a review that finds nothing is indistinguishable from a review that did not look.
Debate Methodology as a Validation Protocol
4. Which modern protocol repairs that weakness, and how?
A test suite. Its detector is mechanical and runs on every change, so it cannot be satisfied by nobody looking, whereas peer review and code review both can.
The rest of this subject
These notes are cut from the University's printed syllabus. Open the syllabus itself for the same subject.