Catalog of Failure Modes

§18.1 The eighteen failure modes

Named so they can be called out in review. Each entry states the failure and its detection test. A list of the opponent's errors only is not a list of errors, so several of these are ones this document has itself committed in draft and had to withdraw — CF-02, CF-05, CF-06, and CF-08 each carry a withdrawal notice for that reason.

1. M01 · Flattening. One register produced and presented as the meaning. Test: does the analysis mention register at all? If not, it has assumed one.
2. M02 · Inflating. Multi-layer treatment of single-register material. Test: can the deeper reading be derived from the text's own vocabulary? If not, it was imported.
3. M03 · Skipping S3. Escalating to the higher register where the ordinary referent bears all predicates. Test: was the S2 test run and recorded? Is the halt condition documented?
4. M04 · The undefeatable reading. No stated condition under which it would be wrong. Test: ask the author what would falsify it. Silence is the finding.
5. M05 · Circular blocking. Invoking lakṣaṇā on grounds that presuppose the conclusion (§9.3). Test: does L1 rest on the disputed thesis? If yes, refuse.
6. M06 · Samākhyā argument. Reasoning from a section's traditional title. Test: rank six of six. Look for a rank-one statement; if none exists, say so.
7. M07 · Decorative derivation. Multiple derivations offered where only one is grammatically valid. Test: check each against N1. Ungrammatical derivations are cut, not softened.
8. M08 · Unlocated citation. A quotation with no traceable source, or a locator that does not resolve — including supplying a plausible-sounding source for an untraced passage. Test: open the printed edition. This test is not negotiable.
9. M09 · Convenience bias. Congenial citations accepted at a lower standard than inconvenient ones. Test: rule C-03. Compare the grading of the last three supporting citations with the last three opposing ones.
10. M10 · Guṇa dismissal. Rejecting a passage by asserting its mode rather than by demonstrating contradiction. Test: rule C-02. Is a specific contradiction with T1–T4 named?
11. M11 · Straw pūrvapakṣa. The opposing view stated in a form nobody holds. Test: can it be sourced to someone who actually holds it, in their words?
12. M12 · Adhikāra as rebuttal. Answering a technical objection by questioning the objector's standing (§4.2). Test: is the objection internal? Then it must be answered on the merits.
13. M13 · Tier inversion. Establishing a tattva conclusion from Purāṇic narrative and then confirming it from śruti. Test: read the argument's order of operations. C-01 fixes the direction.
14. M14 · Cross-sampradāya scavenging. Importing another school's conclusion while discarding the ontology that generated it (CF-05). Test: is the source school named, its premises identified, the conclusion re-derived?
15. M15 · Retrofitting. Reading a modern scientific result back into a verse. Test: rule S3. Would the reading have been produced before the result was known?
16. M16 · Lexical smuggling. Translating a technical term with a modern term carrying alien commitments. Test: rule S4. Check every rendered technical term against its semantic field in the source.
17. M17 · Layer merging. Presenting commentary as though it were mūla, or one commentator's view as the tradition's. Test: is every interpretive statement attributed to a named author at a distinct address?
18. M18 · Manufactured resolution. Producing a settlement for a genuinely unresolved question because an unresolved entry looks weak. Test: check the open register. A method that never returns null is not measuring anything.

§18.2 Automated lint rules and prevention

Several of the eighteen are mechanically detectable and are enforced at build time rather than in review. The <Verse> component requires a grade, so an ungraded citation is a build error rather than an oversight (M08 partially). The <CaseFile> component throws if the residue slot is empty, so a case file with no stated residue cannot ship (M04, M18). Prohibited colour literals, unresolved URNs, and heading-order violations are caught by AST lint.

The remainder — M09, M11, M13 in particular — are not mechanically detectable and are caught only by review against the rules named above. That asymmetry is worth stating rather than hiding: the failure modes that most damage a document of this kind are exactly the ones a linter cannot see.