Standard Context-Control Evidence Assessment
$12,500
One customer-supplied workload, three paths, a written recommendation.
Context-control evidence assessmentFixed scope · inspectable method
Measure one consequential AI workflow before you commit to a larger change.
Maha runs a bounded, reproducible assessment of context control or governed agent actions and returns evidence a technical, risk, or procurement reviewer can inspect.
This is a decision package: a frozen workload, comparable paths, explicit boundaries, and a written recommendation to proceed, revise, or stop. It is not an implementation promise or a certification badge.
Assessment options
A fixed fee for a bounded decision.
Extended Assessment
$25,000
Multiple workloads or a second gateway configuration, with per-workload findings for each.
Founding design partner · $2,500
Available to the first two signed customers, agreeing in advance to act as a named reference and to permit an anonymized integration note.
This is not a general or negotiable discount. It is a fixed exchange for reference participation, and it closes after two customers.
What the assessment produces
A method a reviewer can challenge, reproduce, or decline.
- • A customer-supplied, sanitized document or RAG workload. No production credentials and no personal data.
- • Configuration and workload frozen and digest-recorded before anything runs, so the scoring target cannot move after results are seen.
- • Three paths compared: your baseline, your gateway-native compression, and Maha.
- • Token and cost measurement, evidence retention, citations and provenance, latency, and failure-path behaviour.
- • Sanitized per-workload findings and a written proceed, revise, or stop recommendation.
What to judge Maha on
Controls that remain inspectable after the demo.
- • Deterministic selection: the same inputs produce the same pack, with no model in the path.
- • Hard budgets: the declared token budget is enforced rather than advised.
- • Source-linked provenance: every retained passage carries its source and passage identifier.
- • Stable hashes: input and output commitments a reviewer can recompute.
- • Reproducible evidence: per-workload rows and a one-command check, not a headline.
No retention-superiority claim is made here. The public evidence package includes a dense baseline that scores higher on evidence retention than Maha's production scorer on the frozen MCRB-1 cohort.
Explicit limits
The assessment narrows uncertainty; it does not erase it.
- • No production deployment. The assessment measures; it does not install.
- • No performance or savings guarantee. Nothing is promised before measurement.
- • No certification or compliance opinion of any kind.
- • No open-ended discovery, data migration, or custom implementation work.