Limitations · எல்லைகள்

What this method cannot do.

Every defensible thesis publishes its own limits. These are the limits TLTE knows about, consolidated from across the site so a reviewer does not have to assemble them.

1 · What the method CANNOT do

·Substitute for OHCHR, UN Working Group on Enforced Disappearances, ICJ, or any accountability body. The Mirror-Publish Protocol is precisely the refusal of this role.
·Adjudicate guilt or innocence — TLTE never produces verdicts. It is the citation route to bodies that can.
·Provide emergency safety or safeguarding service. Every survivor-facing page routes to UK 999 / Refuge / SL WIN / TN 181 / Access Now / Citizen Lab.
·Produce its own original counts. Counts cited on the site always carry the originating Tier-A source and the date of that source.
·Speak in a sovereign Tamil voice. TLTE speaks only for Council members who accept the Charter.

2 · Known corpus gaps (tracked)

·Pre-1948 plantation labour archive — uneven coverage outside CWC + 2003 grant materials.
·Eastern Province intra-Tamil + Tamil-Muslim documentation under-represented relative to Northern Province.
·Tamil Nadu fisher livelihood data is uneven between districts (Rameswaram vs Pudukkottai vs Nagapattinam).
·Sinhala-language Tier-A counter-sources are present but under-indexed; coverage tracked at /research/language-coverage.
·Up-country Tamil women's testimony is structurally protected by the Magalir Avai refusal of intake — citations route to PEARL/ITJP, not to the site.

3 · Velicham benchmark blind spots

·Velicham now retrieves bundled markdown corpus passages in production alongside citation glosses; benchmark coverage still has to prove that cultural-depth answers remain citation-locked and do not drift.
·Eval cases are author-curated. Adversarial red-teaming by an independent party has not yet been commissioned.
·Multilingual eval cases (Tamil, Sinhala, French, Arabic) are sparse compared to English cases.
·Long-context cases (>4k tokens of corpus) are under-represented; failure modes there are likely but unquantified.

4 · Resource and funding constraints

·Single-author authorship in era Aarambam — bus-factor of 1. Continuity Protocol addresses succession but has not been tested.
·No grant funding, no donations, no commercial revenue. Infrastructure costs are paid personally.
·No paid legal counsel on retainer. Gate 3 (independent legal review) is currently OPEN.
·No paid DPO. Gate 1 is DESIGNATED interim founder-DPO under UK GDPR Art. 37(2) micro-org provision.

5 · Methodological limits

·The three protocols are presented as named objects, not yet as a formal TLA+ / Alloy model. /research/specification gives the next level down.
·Comparative validation at /research/comparative is honest about transferability but is not a full empirical study of cross-corpus performance.
·The 'two-layer rule' is non-removable as a publishing discipline but is not yet a formally verified property of every page — manual review remains the enforcement mechanism.
·The Continuity Changelog is append-only by editorial discipline; cryptographic hash-chain enforcement is partial (see /continuity/integrity).

6 · Structural choices that are limits, not bugs

·No survivor intake. This is a permanent design choice unless the six Graduation Gates close.
·No naming of survivors, perpetrators, serving officers, at-risk journalists, or families of the disappeared. Permanent.
·No own-voice count aggregation. Permanent.
·No charity status. Permanent.
·No leadership page. Permanent. (Roles change; no role becomes a person.)

If you discover a limit not listed here: write to research@tlte.cloud. Limits named by external readers are added with the date and (where consented) the reader's attribution — this page is itself append-only by editorial discipline.