Technology readiness is not evidence readiness

Evidence audit for deep-tech product development

Request an auditSee a sample audit

OutputOne claim. One dated page.
TurnaroundFive working days.
StandardEvery line checkable against the record.
01

Burden of evidence

Progress is not a matter of accumulating evidence, but of clearing the right evidence.

Every claim in product development activates a burden, one that the development cycle must clear. NADIR audits burdens, records whether they have been cleared, and shows which ones remain unresolved and block progression.

Conventional readiness levels grade activity: what has been demonstrated, in what environment. They do not grade entitlement: the highest claim the record can carry. A device can realistically be at TRL 5 with a measurement never validated in the population it is sold for, and the next study will be spent on it.

A diligence report tells you what the team did and whether it looked competent. It does not tell you which piece of evidence is missing, what it would cost to obtain, or whether the study in the plan addresses anything that blocks the next level.

02

NADIR

Most instruments read evidence backwards. NADIR reads it forward.

Theoretical background

NADIR was developed in cardiovascular medicine, on two large trials of the same hypothesis that returned conflicting results. It explained the conflict: one trial cleared every burden, the other retained two, a population mismatch and a restricted construct. Other instruments name these as indirectness or applicability; none records them as a retained burden attached to a claim and a level.

Most frameworks and appraisal instruments read evidence backwards: once the results are in, what do they stand for? NADIR inverts this approach.

Reading the evidence forward

NADIR reads evidence forward. Before any measurement is taken, it asks which inferential burdens a development will activate. Twenty pre-defined dimensions cover design and positioning, measurement and analysis. Any evidence that meets the standard clears a burden; study designs are not ranked.

Reliability

Two independent readers reach the same result once the claim is fixed (agreement [κ] on [n] claims). Adjudication and the evidence trail are deposited with each reading. This is why NADIR works on claims rather than studies: a product can make several claims, each with its own burdens and its own route to clearance.

Published

[NADIR instrument paper, journal, year, DOI] · [NASA readiness demonstration, deposit, DOI] · [NADIR-TRL framework paper, DOI]

03

NADIR-TRL

A readiness level says what the team has done. NADIR-TRL says what the evidence licenses.

We read the record at any stage of development, settle the level the evidence supports, and audit whether the planned evidence is sufficient to answer. We return one page, dated, checkable line by line.

Two numbers a diligence report doesn't give you

Licensed level. The highest rung whose burdens are all cleared, reported beside the level the company claims. The gap is the first line of the page.

Unlock value. For each unresolved burden, the rung the product would reach if that burden alone were cleared. Often a €10k mapping exercise is worth three levels. Sometimes the planned €500k study addresses nothing that blocks the next one.

What else is on the page

Evidence the company already holds and is not credited for. Whether the next planned purchase addresses any blocking burden. How many attempts each unresolved burden has already absorbed, at what cost, and whether it moved.

04

Four readings

Four readings: two from public NASA development records, one from published cardiovascular trials, and one set locked before results exist. The first three are retrospective: they reproduce the programme's own level where the claim matched and diverge, with the reason, where it did not. The fourth is forward. Each reading is written as PROJECT.claim @ date = { retained | open || cleared }, with the NADIR burden codes; Nd is the retained set alone. The licensed level is extracted from the reading, not assigned. Every cell is sourced; the full adjudication is deposited.

01Activity without movement
Electrochromic space-suit radiatorNASA Game Changing Development · public record

Claim: a full-suit variable-emissivity radiator can actively regulate heat rejection during spacewalks.

RADIATOR.heat-rejection @ 2015-08
= { ∅ | A1 A9 B1·p B3 || A3 A4 A7 A8 A10 B1·a C2 C6 } · Nd = ∅ · Licensed 3 · NASA TRL 3

RADIATOR.heat-rejection @ 2016-01
= { B1·p | A1 B3 || A3 A4 A7 A8 A9 A10 B1·a C2 C6 } · Nd = {B1·p} · Licensed 3 · NASA TRL 3

Binding burden: B1·p, repeatability of the emissivity measurement (level 4). Six months of documented work moved five burdens from open to cleared. The licensed level did not move, because the one burden that controls level 4 was never discharged. A readiness label cannot show this; the ledger shows nothing else. [Deposit: NASA readiness demonstration, DOI]

02One label, two claims
Ejecta STORM lunar-plume instrumentSame hardware · same date · two claims

Claim A: the instrument integrates with a lander and returns usable data under flight conditions.
Claim B: the instrument recovers the entire particle-size distribution of the plume.

STORM.integration @ 2024-02
= { ∅ | ∅ || A3 A4 A7 A8 A9 A10 B1·a B1·p B3 C6 } · Nd = ∅ · Licensed 6 · NASA TRL 6

STORM.full-distribution @ 2024-02
= { A9 | A1 C2 || A3 A4 A7 A8 A10 B1·a B1·p B3 C6 } · Nd = {A9} · Licensed 4 · NASA TRL 6

Binding burden for claim B: A9, restricted outcome range (level 5) — four lasers constrain only part of the distribution. NASA's TRL 6 is correct for the first claim and not for the second. The ledger does not average them. A buyer of claim B is two levels further from market than the label says. [Deposit: NASA readiness demonstration, DOI]

03Premature level evidence
Anti-inflammatory therapy for atherosclerosisCardiovascular medicine · two published trials

Claim: lowering vascular inflammation reduces cardiovascular events in patients with prior events and residual inflammatory risk. CANTOS tested canakinumab; CIRT tested low-dose methotrexate. Population burdens are active here: this is a claim about patients, read through a measurement.

CANTOS.events @ 2017-08
= { ∅ | ∅ || A2 A3 A4 A7 A8 A9 A10 B1·a B1·p C3 C6 } · Nd = ∅ · Licensed 8 · Claim licensed

CIRT.events @ registration 2012
= { A2 | A8 A9 C3 || A3 A4 A7 A10 B1·a B1·p } · Nd = {A2} · Licensed 4 · Level-8 record commissioned

CIRT.events @ 2018-11
= { A2 A8 | ∅ || A3 A4 A7 A9 A10 B1·a B1·p C3 } · Nd = {A2, A8} · Licensed 4

Binding burden at registration: A2, population mismatch. CIRT enrolled on diabetes or metabolic syndrome as a proxy for inflammatory risk; the claim names inflammation, and the record at registration did not carry the marker. That burden was retained before a patient was enrolled and is the first member of the terminal set. A8, exposure contrast, was carried at registration on rheumatology evidence and failed at interpretation — a second route to the same mismatch. The CIRT evidence is not wrong. A level-8 record was commissioned over an open level-5 burden that was visible in the registration, and that is what the ledger reports. [Deposit: registration-epoch profiles and terminal readings, DOI]

04Read before the result
Six inflammation trials · registration epochThree without a result at the time of reading

Six cardiovascular trials share the language of inflammation. Their registrations were read as six bounded claims and profiled on every NADIR dimension at the registration epoch. CIRT is the only one of the six carrying a retained burden before its result existed.

HERMES.events @ registration = { [Nd] | [open] || [cleared] } · Licensed [n] · locked [date]

ATHENA.symptoms @ registration = { [Nd] | [open] || [cleared] } · Licensed [n] · locked [date]

ARTEMIS.events @ registration = { [Nd] | [open] || [cleared] } · Licensed [n] · locked [date]

These profiles are deposited, dated and unchanged. When the results arrive, they either fill the locked claim or they do not; the reading cannot be revised to fit them. Separately, fifty registrations drawn at random from a 1,482-registration frame were profiled on registration alone: the seven most burdened produced no detectable output, the seven least burdened supplied four adjudicable outcomes. A ten-patient open-label trial whose residual burden named several failure modes was followed by a large placebo-controlled trial that met the same modes on the same endpoint. None of this is prediction with a confidence interval. It is a warning issued before the result, on the record, and a result can now falsify it. [Deposit: six-trial overlap audit and registration profiles, DOI] · [Deposit: CTIS pilot, DOI]

Code key: A1 information adequacy · A2 population · A3 temporal alignment · A4 historical baseline · A7 construct–measure mismatch · A8 exposure contrast · A9 restricted outcome range · A10 prior pathway · B1·a instrument accuracy · B1·p instrument precision · B3 temporal instability · C2 model specification · C3 analysis set · C6 mechanism interrogation. Burdens not listed in a reading are inactive for that claim, with the reason recorded in the deposited adjudication.

Before a tranche, before a grant

For a fund: one reading per claim, before the tranche that pays for the next study. The page tells you whether that study addresses a blocking burden, and if not, what would.

For a grant or programme office: the same reading run on a proposal before award. The ladder is the NASA nine-rung scale; a mapping to the European Commission definitions at levels 4 to 6 is available on request and is stated in every reading.

For the company: the ledger names evidence you already hold and are not credited for, and the shortest route to the next level. It is not a score and it is not a verdict on the team.

Delivery

Five working days from adequate access to the record. Adequate means the documents behind the claim, not a summary of them; a reading of a summary is reported as a reading of a summary. Every line points to a document you can check the same afternoon.

What it is not

Not a prediction of success. Not a score. Not a verdict on the team. It covers efficacy and performance claims; harms, commercial viability and engineering feasibility remain with your other diligence. A product with several claims has several levels; they are not averaged.

05

Who reads

A dated instrument, not a negotiated opinion.

[Portrait of the reader]

[Name], [title], [institution]. Epidemiologist and trialist; developed NADIR on cardiovascular outcome trials and ran the readiness demonstration on NASA development records. [One line of track record.]

Independence

A reading is fixed and dated before any clearance work is offered, and clearance work does not re-code it. The company sees the ledger and may add evidence to it; it does not negotiate the coding. Where a reading leads to a Compact, the reader who sealed the ledger is not the one who audits its clearance. Readings are double-coded; disagreements are resolved on the record and the resolution is deposited with the reading.

Three parties

The record belongs to the company. The company grants access, sees the reading in full, and may add evidence before it is dated. The fund receives the same page. A reading states what the record supports on a named claim at a named date; it is not a rating of the company, not a recommendation to invest, and not for circulation beyond the parties who commissioned it.

Confidentiality

The record stays with the reader for the engagement and is not retained afterwards without written agreement. We sign the fund's and the company's confidentiality terms. The adjudication behind the page is available to the commissioning parties and to no one else.

06

Price

Priced by the evidence problem, not the company.

Reading
€3,000–7,000
one claim, one page

One claim, one dated page. The figure depends on how many dimensions the claim activates: a bench measurement claim activates fewer than a clinical outcome claim.

Clearance
per burden
three to five burdens

for each, the cheapest evidence that meets the standard, the design that produces it, and the check that it did. We specify the study; the company runs it. Priced per burden on the same basis; further work by the hour.

Cycle
from €15,000
one product, one cycle

One product, one development cycle, [n] re-readings a year: the ledger re-dated at each purchase decision, with the planned study checked against the open burdens before it is bought. Work beyond the re-readings is by the hour.

Every price is set by two numbers you already know from the reading: claims and dimensions activated. Nothing is charged per company.

07

Sample reading

What one page looks like.

Illustrative output for a single claim: the level the company states, the level the record licenses, and the one burden that separates them.

Claim-level readingIllustrative

Claim: [one sentence, as the company states it]

Claimed
6
On the development plan
Licensed
4
What the record carries
Unlock
+2
If A9 alone were cleared

Binding burden: A9, restricted outcome range. Clear this burden and the evidence can move. Every other line of the page names a document and a date.

Contact

One company.
One week.
One reading you can check.

Give us a portfolio company you are unsure about. We return the reading in five working days; you check it against the record; you decide whether it is worth using on the next one.

Or write directly: [email]