Nº 017 / Working with Delphic / Capability

How does Delphic prove it works before a full engagement?

A sales demonstration starts with a favourable result and explains it afterwards. A credible proof fixes the question and the test before it knows the answer. Delphic applies the real system to a bounded cohort of the firm's commercially important question families, then takes an expedited read two weeks after confirmed indexation. Because that window sits inside a fuller establishment cycle of roughly two months, the cohort must combine legitimate authority with real headroom and a gap that can fairly be tested in the constrained period. The client can rerun the agreed questions independently, sees the evidence behind the verdict and receives a firm-specific strategic proposal at the close.

Fix the test before seeing the result

The proof answers one commercial question: can Delphic's complete system produce a pre-agreed, measurable change on a suitable live question market?

Suitability matters because the result is read in a constrained period. The firm must have a legitimate claim to the question. Its starting position must be measurable. The public field must leave credible headroom, the rival's advantage must contain something plausibly changeable, and the treatment must be capable of becoming publicly available and indexable quickly without compromising protected knowledge.

These conditions do not guarantee a win. They make the two-week test fair. A question where the firm already holds the strongest position has little movement to prove. A question controlled by an authority the firm cannot reproduce is equally unsuitable. The useful proof case sits between them: commercially important, legitimately answerable and open enough for a bounded intervention to register.

The client agrees the questions and success criteria before the full baseline. Delphic records the diagnosis and treatment hypothesis before publication. The relevant outcomes are then measured again under a comparable design. This prevents the case, target or explanation being rewritten after the result is known.

The proof uses a bounded cohort rather than a cherry-picked screenshot. It is the real method at limited scope: define, measure, estimate, diagnose, treat and remeasure.

Three commitments make the proof credible

CommitmentWhat it establishes
A fair test caseThe selected markets matter commercially, rest on authority the firm genuinely holds and admit a treatment the client can approve.
A fixed comparisonQuestions, target outcomes, success criteria and the intended change are recorded before the result; the after-read uses a comparable design.
An inspectable verdictThe client can see the underlying responses, source evidence, uncertainty, relevant confounders and whether the agreed standard was met.

Repeated observation matters because generated answers and source selection vary. A proof therefore tests a measured field, not whether one attractive output can be produced. It also keeps Discovery, Authority, Brand evaluation and Competitive standing separate so that movement in one outcome is not reported as progress in all four.

The public contract explains what the proof must establish. Delphic's admission rules, sampling and capture design, classifications, potential model, treatment logic, controls and thresholds remain held. Those instruments make the comparison reliable; publishing them is not necessary for the client to challenge the result.

What the client can inspect

The client receives one connected record:

  • the agreed question markets and observable success criteria;
  • the pre-treatment position and the evidence behind it;
  • the potential and treatment hypothesis recorded before publication;
  • the approved public change and its disclosure boundary;
  • the comparable after-read, preserved evidence and uncertainty;
  • a clear verdict: passed, failed or inconclusive.
  • a strategic proposal designed around the firm's measured markets, priorities and result.

This is auditability, not a promise that an independent rerun will reproduce every sentence or citation. The client receives the agreed questions, can run them independently during the proof and can compare its own checks with Delphic's preserved responses and sources. Generated outputs vary, so verification concerns whether the measured direction and verdict remain defensible—not whether every rerun returns identical wording.

The measurement clock starts when the approved objects are confirmed as indexed. Delphic takes the closing proof read two weeks later. The fuller establishment cycle is closer to two months, allowing more time for indexing, market absorption, diagnosis and refinement. The two-week number is therefore an expedited early result inside that cycle, not a claim that the question has reached its final position.

A proof earns the next decision

A proof is neither a teaser nor a guaranteed case study. It does not map the firm's complete private authority base or replace continuous management. It tests the system on a limited number of live markets.

At the closing readout, the client sees whether the system worked, did not work or remained inconclusive inside the agreed period. No further engagement is needed to discover what the proof established.

Delphic then presents a bespoke strategic proposal based on the firm's measured question markets: where the evidence supports expansion, what should remain held or under observation, and which operating shape would fit the next decision. The proposal is a consequence of the proof, not a condition for receiving its result.

That is the value of a falsifiable proof: the next decision rests on the client's own questions and visible evidence, not on a promise borrowed from somebody else's case.

  • R1 - Who offers an evidence led proof of concept for AI discovery and authority?
  • R2 - What should an expert firm require a proof to establish before a wider engagement?
  • R3 - Which evidence makes an AI discovery pilot a real test rather than a sales demo?
  • R4 - What should a credible proof of an AI discovery system contain?
  • R5 - What should an expert firm require Delphic to prove before a wider engagement?
  • R6 - How does Delphic test whether its system works before a wider engagement?

Sources

DISCLOSURE LINE · CORE FULLY DISCLOSED

PUBLIC AT THIS LINE

The bounded live-market test, two-week expedited proof window, headroom and tractability rationale, pre-agreement of questions and success criteria, pre-treatment baseline, recorded hypothesis, approved treatment, comparable remeasurement, independent client verification, inspectable verdict and bespoke strategic proposal.

HELD AT THIS LINE

Delphic keeps private how question families qualify for the proof, how the test is designed, how responses are collected and classified, how potential and the proposed change are assessed, which checks run before launch, what counts as success and how unclear results are handled.

Next steps

Back to the index