Compressed Air Project Evidence Centre

Compressed Air Project Evidence Centre

1. The answer: what qualifies as project evidence?

A compressed-air project belongs in an evidence centre only when the published statement can be traced to an identifiable project, a defined engineering boundary, dated original records, an appropriate measurement or acceptance method, a responsible reviewer and publication permission. A project photograph alone establishes an observed configuration at most; it does not prove energy savings, reliable operation, customer satisfaction or compliance. The evidence centre is a controlled index of claims and their supporting records, not a gallery of unverified success stories.

Preserve those boundaries when writing introductions and anchor links.

2. Evidence tiers and permissible claims

Evidence tiers are an internal editorial classification, not a formal certification or a claim that any third party has verified a project. A higher tier cannot be inferred from the presence of many files: completeness, comparability, permission and technical validity matter more than volume.

3. The six evidence questions every project card must answer

4. Identity gate

Which customer or anonymised facility, project ID, asset ID, location granularity and project period are supported? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

5. Intervention gate

What was actually changed, installed, measured, repaired or commissioned, and what was explicitly excluded? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

6. Baseline gate

What pre-intervention condition is evidenced, at what pressure, production rate, shift pattern and instrument configuration? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

7. Outcome gate

Which observation, measured delta or accepted scope can be stated without attributing unsupported causation? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

8. Permission gate

Who owns the data, photos, trademarks and site identifiers, and what use was authorised, for how long and on which channels? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

9. Review gate

Who signed the technical claim, uncertainty treatment, editorial wording, confidentiality redaction and publication decision? The catalogue must store both the conclusion and a direct reference to the supporting record. Where a record is unavailable, mark the item PENDING or NOT PUBLISHABLE rather than using an implied approval. A reviewer must be able to reconstruct the decision without relying on the author’s memory.

10. Project evidence architecture: a record is not a story

A project is a parent entity; its evidence objects are versioned children. Separate the scope record, technical baseline, intervention, measurements, acceptance, permissions and publication claim. One project can support several claims, but each claim must have its own evidence references and a distinct boundary. For example, a new compressor installation may have a verified commissioning record without having any defensible electricity-savings figure. Publishing the installation is permissible if consented; adding an efficiency claim is not.

Recommended catalogue keys: project_id; evidence_id; claim_id; revision; category; event_date; measurement_period; source_owner; consent_scope; evidence_tier; confidentiality_class; technical_reviewer; editorial_reviewer; publication_status; expiry_or_review_date. Preserve stable IDs when records are corrected. Public URLs should not expose internal document IDs or customer-sensitive identifiers.

11. Catalogue structure and content model

12. Project categories and eligibility

The centre may organise approved entries by leakage repair, sequencing optimisation, heat recovery, compressor installation, overhaul, temporary-air deployment, treatment retrofit, piping modifications and maintenance improvements. These are navigation categories, not assertions that a project exists. A category remains hidden or displays an explicitly non-promotional evidence-submission explanation until an eligible record is approved. Never populate a category with invented logos, before-and-after photos, project names or dates.

Category landing cards should show a concise verified scope, industry only if consented, evidence tier, measurement basis where relevant, last review date and a link to the full record. Avoid headline savings percentages without denominator, period and uncertainty. Do not equate a contractor invoice with proof of an engineering outcome.

13. Evidence ingestion workflow

Intake

Create a stable project ID; receive files via access-controlled storage; log source, owner, original dates and collection purpose. Flag confidential material before indexing. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Identity reconciliation

Cross-check site name, work order, asset serial, job card, acceptance record and photo metadata. Resolve contradictory dates or equipment IDs before creating a public record. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Claim decomposition

Rewrite promotional copy into atomic claims: installed X, measured Y, observed Z. Each claim receives its own evidence ID and review decision. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Technical screening

Check pressure and flow reference conditions, units, instrument calibration, test duration, system boundaries, representative demand and known confounders. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Permission screening

Obtain documented rights for customer identity, logo, photographs, employee quotes, project details and distribution channels. Consent to perform work is not consent to publicise it. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Publication preparation

Create an approved summary, anonymised metadata, accessible images and clearly bounded numbers; reconcile all links and structured data with visible content. A failure or unknown at a mandatory gate is a hold, not a pass by default.

Release and surveillance

Publish only after technical, editorial, privacy and staging approval; monitor source expiry, corrections, consent withdrawal and changes to the claim. A failure or unknown at a mandatory gate is a hold, not a pass by default.

14. Measurement boundary and baseline design

A defensible before/after comparison identifies where electrical input was metered, whether air flow was delivered or inferred, pressure at the measurement location, treatment state, operating hours, production demand and weather or ambient effects where material. Compare like-for-like operating periods or use a documented normalisation method. If the system boundary changes between periods, state it prominently; a compressor-specific measurement must not be presented as whole-station performance.

For energy reporting, distinguish instantaneous kW from cumulative kWh, measured savings from projected annualisation, and energy intensity (kWh per unit of output) from absolute consumption. A lower electricity bill can reflect tariffs, fewer production shifts or a shutdown. It is not by itself evidence of compressor optimisation.

15. Uncertainty and causation: publish the narrower claim

For two independent measurements x1 and x2, the observed difference is Δ=x1−x2. When the individual standard uncertainties u1 and u2 are independent, a first-order combined standard uncertainty is uΔ=sqrt(u1²+u2²). Correlated instruments, common calibration biases and model assumptions require a different treatment; do not automatically use the independent formula. If a result is comparable to uncertainty, avoid a categorical improvement claim. State the observed change, conditions and limits.

Causation requires stronger evidence than chronological order. A pressure reduction after a repair might coincide with production downtime, controller retuning or changed demand. State “measured during the test period” when an isolated intervention effect has not been demonstrated. Persistence requires subsequent observations; a single commissioning test cannot support a year-long reliability claim.

16. Worked case A — leakage repair claim, illustrative only

Assume an anonymised station records estimated leak-related air demand of 2.1 m³/min before repairs and 1.3 m³/min afterwards using the same documented method and pressure band. The observed difference is 0.8 m³/min, or 38.1% of the original estimate. This is an illustrative measurement arithmetic example, not a customer outcome. Without demand representativeness, instrument uncertainty, operating hours, compressor response and persistence evidence, it cannot establish annual kWh or currency savings.

Catalogue disposition: E3 candidate, pending method report, pressure evidence, calibrated instrument information and authorised project consent. Public claim: HOLD.

17. Worked case B — electrical reduction and output change

Illustrative data: Period A uses 82,000 kWh for 10,000 units of output; period B uses 75,000 kWh for 8,000 units. Absolute energy fell 7,000 kWh (8.5%), while intensity rose from 8.2 to 9.375 kWh/unit (14.3%). A headline stating “8.5% efficiency improvement” would be misleading because the production denominator changed. The correct conclusion is a reduction in total energy alongside deterioration in measured intensity, subject to boundary and data validation.

Evidence requirement: metering source, output definition, representative production mix, shifts, compressed-air-only allocation, tariff separation and any other changes. Catalogue decision: do not claim compressor efficiency improvement from these figures.

18. Worked case C — pressure stabilisation

Illustrative compressor-room data: pressure at the receiver changes from a pre-project 6.4–7.4 bar(g) band to a post-project 6.8–7.2 bar(g) band over separately logged intervals. The observed span narrows from 1.0 to 0.4 bar. This does not prove that end-use pressure became more stable: simultaneous consumer-side measurements, sensor accuracy and demand comparability are needed. A receiver trend cannot substitute for the most remote critical-use pressure.

Publication decision: a bounded receiver-pressure observation might be supported at E3 if the original time series exists. Claims of reduced defects, improved production or energy savings require additional evidence.

19. Worked case D — heat recovery

Illustrative measurements: water flow 1.2 kg/s, water temperature rise 7 K, assumed specific heat 4.18 kJ/(kg·K). Thermal transfer estimate Q=1.2×4.18×7=35.1 kW. This describes estimated heat delivered across a measured water boundary at that moment, not annual useful heat or avoided fuel. Demonstrating actual benefit additionally requires sink demand, heat exchanger losses, operating hours, displaced heating source efficiency and control interactions.

Catalogue decision: no annual carbon or currency figure without period data and a documented counterfactual.

20. Worked case E — installation acceptance versus performance

An illustrative handover packet contains a purchase order, installed-machine photographs, piping pressure-test record and commissioning signature. Those items may support an installation scope statement, provided asset identity, date and permission match. They do not prove “zero breakdowns”, “guaranteed 24/7 availability” or reduced operating costs. Claims must be split: executed work (E2 candidate) versus unverified future reliability (E0).

21. Worked case F — overhaul before/after

An illustrative overhaul record has as-found vibration 6.2 mm/s RMS and as-left 3.4 mm/s RMS, measured at a stated bearing location under similar speed/load conditions. The numerical reduction is 2.8 mm/s RMS, but acceptance depends on OEM and machine-specific vibration limits, sensor location, frequency content and operating state. A single scalar value does not prove remaining bearing life or full machine reliability. Technical sign-off must precede a public performance claim.

22. Worked case G — uncertainty illustration

Assume two independent test values of 100±3 kW and 96±3 kW, with ±3 kW treated as standard uncertainty for this teaching example. Difference=4 kW; combined standard uncertainty=sqrt(3²+3²)=4.24 kW. The difference is smaller than one combined standard uncertainty. It would be inappropriate to present the 4% numerical difference as a conclusively demonstrated efficiency gain. Confirm uncertainty definitions before real use.

23. Worked case H — consent and redaction

A fictional project folder includes an authorised job description but no permission for the customer logo, site photograph or employee testimonial. The project team may internally assess technical evidence, but the public card must omit those assets until the rights owner grants channel-specific permission. Anonymisation must also remove revealing landmarks, file metadata and identifiable security details. Consent status is independent of technical validity.

24. Mandatory asset — Evidence Catalogue with Source and Consent Records

Each field must permit a source reference and a status: VERIFIED, PENDING, NOT APPLICABLE WITH REASON, REJECTED or WITHDRAWN. An empty field is not equivalent to “not applicable”. Maintain an internal complete record and a public-safe projection. Only approved claim text is eligible for website rendering.

25. Example catalogue entry — deliberately fictional

26. Public page information architecture

The /evidence/ route should begin with a precise explanation of the evidence criteria and an accessible category navigation. Verified project cards, when available, show title, limited industry context, what was measured, evidence tier, period, limitations and the approved case URL. A clear “how we verify” section links to measurement and consent methodology. Empty categories should not be padded with fabricated case studies. Do not display zero-result carousels as proof of missing capability.

Search filters may use project category, evidence tier, industry and documented technology. Filters must be driven from approved metadata, keyboard accessible and correctly announce result counts. A project can be removed from public results without deleting the controlled internal evidence history. Protect confidential originals behind access control; never embed raw internal documents in HTML, JSON-LD, image EXIF or public file URLs.

27. Image, video and testimonial integrity

For every photo or video, store the original file, capture date where known, rights owner, permission scope, depicted asset, any editing performed and an accessible text alternative. A generic image must be labelled representative rather than assigned to a specific customer. A staged or AI-generated image must never be represented as a real site record. For testimonials, retain original wording, attributable permission, speaker role and any commercial relationship. Do not create customer quotes to fill the page.

28. Correction, withdrawal and retention

When a claim is disputed, unpublish or qualify it according to risk while retaining the controlled audit trail. Log original statement, reason for correction, evidence affected, reviewer, decision date, affected URLs, caches and structured data. Permission withdrawal must trigger removal of covered images, names, quotes and derivative snippets where controllable. Re-check pages after project scope, instrument calibration, consent or underlying data changes. Define a retention period through company policy and applicable law, not a fabricated universal duration.