Freeze before scoring
Source rights, baseline, primary metric, threshold, holdout, failures, and allowed claims are named first.
30-day bounded validation sprint
One candidate. One accepted baseline. One decision you can defend.
LumenCore helps a technical decision owner determine whether a promising AI, forecast, routing, optimization, or infrastructure candidate should be promoted, rerun, held, rejected, or sent to external review—under rules fixed before scoring.
The customer problem
A team has a promising result, but procurement, a funder, or a technical reviewer cannot tell whether it beats the incumbent under locked rules—or whether the evidence was tuned, filtered, or promoted after the outcome was known.
Run matched conditions
Candidate and baseline use the same authorized source and evaluation contract. Neutral and failed results stay visible.
Return one bounded decision
The output is promote, rerun, external review, hold, or reject—not a broad promise that the technology works everywhere.
Frozen electricity-forecast comparisons · September 2026
Inspect the results and the gates they did not pass.
Two retrospective windows use the same development-selected candidate. These are first-party replay results. The complete download preserves inputs, predictions, hashes and the original failed gates.
| Window | Paired authority-days | Direct LightGBM baseline | Residual candidate | Error reduction |
|---|---|---|---|---|
| Original historical window | 1,176 | 17,874.172 MWh | 15,979.087 MWh | 10.602% |
| July 13–September 8 archive | 285 | 19,765.544 MWh | 16,107.044 MWh | 18.509% |
Decision: HOLD. Original coverage and a SWPP regression gate failed. The later window has no eligible SWPP rows and too few days and authority-month units. Historical records may contain revisions. These comparisons do not prove prospective performance, independent validation, electricity saved or annual dollar savings.
Verify the download
ZIP SHA-256: cc05eb52559bca53035bfe09e36438ad191634767e2d9cc684f94048d448a8fb
After extraction, run python code/ops/VERIFY_EIA_CONSTRAINT_REVIEW.py --packet original and repeat with --packet later. Hash verification establishes artifact identity; it does not establish scientific correctness or a favorable decision.
The next useful test needs a buyer-owned incumbent with matching target and issue time, complete source history, unchanged service requirements and an agreed operating-cost policy. Discuss one validation scope.
What the buyer receives
Pre-registered contract
Named data rights, comparator, metric, threshold, holdout, failure rules, decision owner, and claim boundary.
Reproducible evidence pack
Input, code, dependency, and output manifests; matched results; negative findings; SHA-256 custody; offline instructions.
Proof Capsule and briefing
A reviewer-readable result that separates measured, replay, synthetic, modeled, and externally validated statements.
AuthorizeDays 0–3
FreezeDays 4–7
ExecuteDays 8–17
ChallengeDays 18–24
DecideDays 25–30
Founder-approved launch scope
Pricing is an offer hypothesis, not booked revenue or guaranteed market value. Final scope, data handling, IP, acceptance, and payment terms belong in the signed agreement.
| Tier | Launch price | Bounded scope | Best fit |
|---|---|---|---|
| Launch Replay | $7,500 | One source, one baseline, one primary metric, one held-out window, one Proof Capsule, one briefing. | Startup, lab, or operator needing one credible decision. |
| Standard Sprint | $15,000 | Up to three held-out windows, one controlled reviewer rerun, offline verifier, technical and executive briefings. | Buyer or reviewer needing repeatability. |
| Institutional Sprint | From $25,000 | Up to two sources and baselines, buyer-owned holdout, two reruns, integration workshop, procurement handoff. | Utility, enterprise, prime contractor, or regulated team. |
ProofLock
Custody is not authority.
ProofLock verifies receipt and artifact identity, derives machine-checkable lineage, and blocks a self-authored receipt from minting baseline, held-out evaluation, independent-review, or human-release approval.
The bundled public demonstration intentionally ends at HOLD.
What it does not prove
Artifact hashes do not establish performance, realized savings, certification, field validation, safety approval, deployment readiness, or buyer acceptance.
Verify the demonstrationStart with three short answers.
Tell us about one real task or decision. A few sentences are enough; write UNKNOWN for anything you have not established.
- Your business and decision: What recent task or technical decision would you like to improve?
- What you use today: What process, tool or baseline would a useful improvement need to beat?
- Owner and timing: Who reviews the result, and when would it be useful?
Send a non-confidential description only. Keep datasets, personal records and account credentials out of the first email. This starts a fit conversation; it does not book a sprint, transfer data rights or accept terms.
If no email app opens, copy the three questions into your email and send them to robertashworth4444@gmail.com. We review fit first; scope, deliverables and commercial terms are agreed separately.
No favorable result is promised. The sprint does not establish guaranteed ROI, realized savings, certification, agency endorsement, award likelihood, autonomous control authority, or field validation. A neutral or failed result remains part of the deliverable and may prevent a more expensive bad decision.