-
Notifications
You must be signed in to change notification settings - Fork 6
Kalama10
Kalama10 is FPCOS's 10-gate proof standard. It is derived from the Kalama Sutta — a discourse on rational inquiry that pre-dates modern epistemology by two millennia — adapted for AI reasoning and claim verification.
The core principle: Accept a claim only when it survives scrutiny from evidence, consequence, and independent verification — not because it sounds right, comes from authority, or fits an existing model.
"Do not go by oral tradition, by lineage of teaching, by hearsay, by a collection of scriptures, by logic, by inferential reasoning, by reasoned cogitation, by the acceptance of a view after pondering it, by the seeming competence of a speaker, or because you think: 'The ascetic is our teacher.' When you know for yourselves: 'These things are unwholesome...' — then you should abandon them." — Kalama Sutta (adapted)
A claim is ACCEPTED only when it passes all 10 gates:
| Gate | Reject if accepted only because | Required instead |
|---|---|---|
| K1 | Rumor / word-of-mouth / "I heard that..." | Independently verifiable source |
| K2 | Tradition / "it's always been done this way" | Evidence of current effectiveness in this context |
| K3 | Logic alone, no empirical grounding | Logic + at least one observed real-world consequence |
| K4 | Analogy alone — "it's like X, and X works" | Direct evidence for this specific case |
| K5 | Preference / "it feels right" / comfort | Preference ≠ truth — label preference as preference |
| K6 | Authority alone — "experts say..." / "studies show..." | Authority + independently testable claim |
| K7 | Theory-fit — confirms existing model | Alternative models seriously considered |
| K8 | Single source / single teacher | Multiple independent sources agreeing |
| K9 | Consensus alone — "everyone agrees..." | Consensus + traceable evidence chain |
| K10 | "Seems reasonable" / plausibility | Operationally tested in reality |
Kalama10 does not require that every claim be proven before it can be used. It requires that the epistemic status of every claim be declared — which gates it has passed, which it has not.
Examples:
"The Fed rate decision will impact gold prices"
- K1: Not rumor — established economic relationship
- K3: Logic confirmed — rate rises → dollar strengthens → gold denominated in USD falls in international demand
- K7: ✅ Alternative model considered — in USD terms, gold also acts as inflation hedge, may rise if inflation fear exceeds rate impact
- Status: ACCEPTED with caveat — direction uncertain, competing mechanisms
"Experts agree that X product will succeed"
- K6: FAILED — authority without independently testable claim
- K8: FAILED — "experts" may be from the same incentive field
- K9: FAILED — consensus without traceable evidence chain
- Status: REJECTED as proof. Reclassify as INFERRED (optimistic model, authority-based)
A claim is not proven until you can explain WHY it is true from first principles — not just that it IS true.
This is the difference between:
- Recall: "Gold falls when the dollar rises." (knows the correlation)
- Understanding: "When the dollar rises, the same amount of gold costs more dollars — reducing demand from non-USD buyers, putting downward pressure on price." (knows the mechanism)
Kalama10 requires understanding, not recall. A claim passed by recall alone fails K3 (logic without mechanism) and K10 (not operationally understood).
Domain expertise does not exempt any claim from Kalama10.
Being an expert narrows your prior — it does not eliminate the need for verification. In fact, experts are most vulnerable to K7 (theory-fit) and K5 (comfort with familiar models). The more deeply expert, the more likely that confirmation of existing models feels like evidence.
The FPCOS Skill Interface Contract explicitly states: "domain expertise ≠ exemption" from Kalama10.
- L1-Axiom-Gate — Where Kalama10 runs in the pipeline
- Ariya4 — The companion problem-frame protocol
- Anti-Hallucination-Map — K6 maps to hallucination source 2 (unverified authority)
# Kalama10 — Proof Standard
[← Back to Home](Home)
Kalama10 is FPCOS's 10-gate proof standard. It is derived from the Kalama Sutta — a discourse on rational inquiry that pre-dates modern epistemology by two millennia — adapted for AI reasoning and claim verification.
The core principle: Accept a claim only when it survives scrutiny from evidence, consequence, and independent verification — not because it sounds right, comes from authority, or fits an existing model.
"Do not go by oral tradition, by lineage of teaching, by hearsay, by a collection of scriptures, by logic, by inferential reasoning, by reasoned cogitation, by the acceptance of a view after pondering it, by the seeming competence of a speaker, or because you think: 'The ascetic is our teacher.' When you know for yourselves: 'These things are unwholesome...' — then you should abandon them." — Kalama Sutta (adapted)
A claim is ACCEPTED only when it passes all 10 gates:
| Gate | Reject if accepted only because | Required instead |
|---|---|---|
| K1 | Rumor / word-of-mouth / "I heard that..." | Independently verifiable source |
| K2 | Tradition / "it's always been done this way" | Evidence of current effectiveness in this context |
| K3 | Logic alone, no empirical grounding | Logic + at least one observed real-world consequence |
| K4 | Analogy alone — "it's like X, and X works" | Direct evidence for this specific case |
| K5 | Preference / "it feels right" / comfort | Preference ≠ truth — label preference as preference |
| K6 | Authority alone — "experts say..." / "studies show..." | Authority + independently testable claim |
| K7 | Theory-fit — confirms existing model | Alternative models seriously considered |
| K8 | Single source / single teacher | Multiple independent sources agreeing |
| K9 | Consensus alone — "everyone agrees..." | Consensus + traceable evidence chain |
| K10 | "Seems reasonable" / plausibility | Operationally tested in reality |
Kalama10 does not require that every claim be proven before it can be used. It requires that the epistemic status of every claim be declared — which gates it has passed, which it has not.
Examples:
"The Fed rate decision will impact gold prices"
- K1: Not rumor — established economic relationship
- K3: Logic confirmed — rate rises → dollar strengthens → gold denominated in USD falls in international demand
- K7: ✅ Alternative model considered — in USD terms, gold also acts as inflation hedge, may rise if inflation fear exceeds rate impact
- Status: ACCEPTED with caveat — direction uncertain, competing mechanisms
"Experts agree that X product will succeed"
- K6: FAILED — authority without independently testable claim
- K8: FAILED — "experts" may be from the same incentive field
- K9: FAILED — consensus without traceable evidence chain
- Status: REJECTED as proof. Reclassify as INFERRED (optimistic model, authority-based)
A claim is not proven until you can explain WHY it is true from first principles — not just that it IS true.
This is the difference between:
- Recall: "Gold falls when the dollar rises." (knows the correlation)
- Understanding: "When the dollar rises, the same amount of gold costs more dollars — reducing demand from non-USD buyers, putting downward pressure on price." (knows the mechanism)
Kalama10 requires understanding, not recall. A claim passed by recall alone fails K3 (logic without mechanism) and K10 (not operationally understood).
Domain expertise does not exempt any claim from Kalama10.
Being an expert narrows your prior — it does not eliminate the need for verification. In fact, experts are most vulnerable to K7 (theory-fit) and K5 (comfort with familiar models). The more deeply expert, the more likely that confirmation of existing models feels like evidence.
The FPCOS Skill Interface Contract explicitly states: "domain expertise ≠ exemption" from Kalama10.
- L1-Axiom-Gate — Where Kalama10 runs in the pipeline
- Ariya4 — The companion problem-frame protocol
- Anti-Hallucination-Map — K6 maps to hallucination source 2 (unverified authority)
[← Back to Home](Home)