Key numbers
Contract proof dispute math (ISO 12647-2 / 12647-7 / ISO/TS 15311-1):
- Proof validity is verified by a measured control target — in practice the Ugra/FOGRA Media Wedge CMYK printed alongside the subject.
- Production color variation is judged by the 95th percentile of ΔE00, not by standard deviation: color differences are not normally distributed.
- Typical variation limits: ΔE00 ≤ 2.0 for C, M, Y; ≤ 3.0 for K; ≤ 1.5 for spot colors (95th percentile against the OK print).
- Solid deviation from aim: ΔL < 5, ΔC < 3, Δh < 6°** (spot colors: Δh < 8°). Midtone stability: 50% tint within ±4% tone value over the run.
- Measure under a defined ISO 13655 M-condition (M1 for OBA papers) or the numbers are not comparable at all.
A client rejects a delivered run: "the color doesn't match the proof." At this moment the dispute is either an engineering conversation about numbers — or a negotiation about who absorbs the reprint. Which one it becomes depends entirely on what the printer measured and logged before the truck left the dock.
The proof is only a contract if it carries a wedge
ISO 12647-2 requires that files delivered for printing be accompanied by a contract proof that simulates the intended printing condition and conforms to ISO 12647-7 — and that this conformance be verifiable by measurement of a control strip printed on the proof itself. In practice that means an Ugra/FOGRA Media Wedge CMYK next to the image.
A proof without a measurable wedge is not evidence. It is a nice picture. If your client's agency supplies proofs without wedges, you have no reference against which "matches / doesn't match" can be objectively decided — and in a dispute, the party without numbers loses by default.
Why the 95th percentile, not the average
When a run is sampled — say 15 sheets against the OK print — intuition says: compute the mean ΔE00 and the standard deviation, like any process metric. ISO/TS 15311-1 explicitly forbids that intuition: color differences do not follow a Gaussian distribution, so standard deviation is a mathematically inappropriate spread metric. The standard requires reporting the mean and the 95th percentile of ΔE00.
This matters commercially. An average of ΔE00 = 1.2 sounds excellent, but if the 95th percentile is 4.5, one sheet in twenty is visibly off — and that is exactly the sheet the client will put on the table. Conversely, a printer who can show CS95 ≤ 2.0 on the control strip has a statistically defensible run even when a single cherry-picked sheet looks worse.
Typical acceptance values used in contracts (aligned with ISO 12647-6 drafts and ISO 15311-1 practice):
| Parameter | Tolerance |
|---|---|
| Variation, C / M / Y (95th pct ΔE00 vs OK print) | ≤ 2.0 |
| Variation, K | ≤ 3.0 |
| Variation, spot colors | ≤ 1.5 |
| Solid deviation from aim | ΔL* < 5, ΔC* < 3, Δh < 6° (spot Δh < 8°) |
| 50% tint over the run | within ±4% tone value |
Measurement conditions decide the argument before it starts
Numbers from two instruments are only comparable when the ISO 13655 M-condition matches. On modern papers loaded with optical brightening agents, a legacy M0 (undefined UV) measurement and an M1 (D50 UV content) measurement of the same sheet disagree — predictably on the b* axis. M2 (UV-cut) has its place only for UV-free viewing environments; M3 (polarized) exists so offset printers can predict dry density from a wet sheet.
If the client's QC measured M0 and you measured M1, both sides are honestly reporting different numbers for the same physical object. Fix the M-condition in the job specification — one line of text that prevents an entire class of disputes.
The data log that ends rejections
To be defensible, the printer's run report needs, at minimum:
- Reference printing condition and profile (e.g., PSO Coated v3 / FOGRA51) — agreed in writing.
- Substrate CIELAB (Lab*, M1) — because aims shift with paper white.
- OK print solid values with deviations from aims (ΔL*, Δa*, Δb*, ΔE00 per primary).
- Production variation: mean and 95th percentile ΔE00 per primary against the OK print, from a stated sample size.
- Midtone TVI per channel and its drift over the run.
- The M-condition and instrument used.
That is one measured strip per sampling interval and a template. Against it, "looks different to me" is no longer a commercial threat.
Self-check: are you dispute-proof today?
- Do your contract proofs carry an Ugra/FOGRA wedge — and do you actually measure it at OK-sheet time?
- Is the M-condition written into your job tickets?
- Can you produce a 95th-percentile ΔE00 for your last disputed run, or only an average?
- Is your sampling plan (how many sheets, at what interval) agreed with the client before the run?
If any answer is "no," the gap is procedural, not technological — and it is exactly the kind of gap an independent audit closes in days, not months.