A Regulatory Change Becomes a Machining Problem

Changing a recycling mark on the bottom of a plastic bottle sounds like a graphics edit. It is not. The mark is formed by the metal surface inside the mold, so changing the bottle means modifying production tooling.

California SB 343 is one reason manufacturers are reassessing molded-in recycling marks. The law restricts the use of chasing arrows and other indicators of recyclability unless the product or packaging meets specified criteria. For rigid plastic bottles and containers, it also addresses when the required resin identification code may appear inside a chasing-arrows symbol.

A resin identification code, or RIC, identifies the type of plastic resin used in a product. It does not, by itself, establish that the product is recyclable. CalRecycle currently identifies October 4, 2026 as the manufacturing date after which the labeling restrictions apply, while also reporting that a federal preliminary injunction issued on July 14, 2026 blocks enforcement while litigation continues.

Regardless of how that litigation develops, manufacturers revising molded-in markings face the same physical problem: the change has to be made in the mold.

We recently faced that problem on a group of blow mold tailstocks—the removable mold components that form the bottoms of bottles. The existing markings had been filled and hand blended into the surrounding cavity surfaces. New RIC markings then had to be engraved into the restored metal.

This was not a batch of identical inserts. The tailstocks varied in overall size, mounting arrangement, utility connections, and local cavity geometry. Each part required its own setup and, as we would learn, its own measured surface model.

At first glance, the job looked like ordinary engraving. The real engineering question was harder: How do you cut a shallow, consistent mark on a curved mold surface when the actual finished shape of that surface is no longer known accurately enough?

Why a Few Thousandths of an Inch Matter

The completed RIC was only 0.346 × 0.426 inch. Its deepest programmed cut was 0.011 inch, made with a 0.031-inch ball end mill. At that scale, a few thousandths of an inch can dramatically change the finished mark.

Consider a nominal engraving depth of 0.010 inch. If the actual mold surface sits 0.003 inch higher than the program expects, the tool cuts 30% deeper than intended. If the surface sits 0.003 inch lower, the cut is 30% shallower.

The rounded tool tip makes the effect more visible. A deeper cut with a ball end mill produces a wider, heavier line. A shallower cut produces a narrower, fainter line and can allow part of the symbol to disappear. The XY geometry may be perfect while the finished mark still looks uneven.

On the representative tailstock used throughout this article, the measured surface height changed by 0.022 inch across the engraving path. That is more than twice the entire 0.011-inch engraving depth. A toolpath held at one fixed Z height could not produce a consistent mark on that surface.

Why the Usual Engraving Approaches Failed

The graphic itself was easy to program. The difficult part was defining where the metal surface actually existed beneath every point of the toolpath.

One Touch-Off Was Not Enough

Our first approach was conventional. We used a Renishaw spindle probe on a Haas machining center to locate the component and establish a representative Z height. We then ran the engraving in a fixed plane.

That approach was not accurate enough. The engraving area looked nearly flat by eye, but “nearly flat” is not the same as flat within a few thousandths of an inch. One side of the symbol could cut correctly while another side became too heavy, too faint, or incomplete.

We also considered dividing the symbol into regions and manually assigning different Z depths. That can improve one part through trial and error, but it depends on operator judgment, creates abrupt changes between regions, and does not scale to a family of individually different tools. We needed continuous compensation, not a collection of manual corrections.

The CAD Model Was Not the Mold

A curved surface is not inherently difficult to engrave when its true geometry is known. With an accurate 3D model, modern CAM software can project a toolpath onto that surface and maintain a controlled depth.

The problem with legacy production tooling is that nominal CAD may no longer describe the finished metal closely enough. Mold surfaces are polished, blended, repaired, filled, and dressed by hand over years of service. Those changes may be completely acceptable for bottle production while still being large relative to a 0.010-inch-deep engraving.

In this project, reliable finished-surface CAD was not available for every tailstock, and one component could not be treated as a substitute for another. The only surface that mattered was the one the cutter would actually encounter.

Probe measurements confirmed what the physical parts suggested. Even tailstocks in the same 89 mm family did not share the same local surface shape.

To compare the local geometry rather than setup position, we mathematically removed each part’s overall height and tilt. At 25 corresponding probe locations across three tailstocks, the average spread was 0.00118 inch and the largest was 0.00327 inch. That maximum difference is roughly one-third of a 0.010-inch engraving depth.

Those measurements cannot tell us exactly how much of the difference came from original manufacturing, later repairs, wear, polishing, or handwork. They do tell us what matters for machining: the cutter would encounter materially different surfaces on nominally similar components.

Machining a Flat Would Solve the Wrong Problem

A small flat pad would make the engraving easy. It would also change the mold surface and therefore the bottle surface.

Removing enough material to establish a flat could create a visible witness, a local thickness change, or a geometric discontinuity in a mold that was already producing acceptable bottles. The objective was not to machine the mold until it matched a simple engraving program. The objective was to modify the engraving program until it followed the mold.

The Solution: Measure the Tool, Then Warp the Program

The capability we developed separates the job into three controlled steps:

1. Measure the mold surface in and around the region the cutter will traverse.
2. Reconstruct a smooth surface from the measured XYZ points and test whether that reconstruction is trustworthy.
3. Rewrite the original CNC program so each cutting move follows the measured surface while preserving the original CAM intent.

The first step is deliberately targeted. We did not need to reverse-engineer the complete tailstock. We needed an accurate local map beneath the engraving and enough surrounding measurements to constrain that map.

The actual production routine used 25 probe points around a cutting footprint only 0.346 × 0.426 inch. Some points lay directly along important portions of the engraving path. Others surrounded the mark or filled gaps between features. The engraving itself helped determine where measurements would be most valuable.

This is an important distinction. Full 3D scanning is valuable when an entire shape must be reconstructed. Here, the machining region was small and the consequence of local error was high. A focused probe map provided the information needed for the cut without turning a localized mold modification into a complete reverse-engineering project.

First, Prove the Measurements Are Trustworthy

Before allowing software to rewrite production G-code, we validated the complete measurement chain: the spindle probe, the recorded coordinates, the geometric calculations, and the surface-fitting logic.

A 5 × 5 map on a flat test surface showed only 0.00040 inch of total point-to-point variation. Its best-fit surface had an RMS residual of 0.000058 inch.

We then measured a known concave cylindrical surface with a physical diameter of 1.600 inches using a 6 mm stylus ball. Because the center of the stylus ball follows a smaller radius than the metal surface, the expected probe-center radius was 0.6819 inch. The radius recovered from the measurements was 0.6807 inch—a difference of 0.00114 inch, or about 0.17%.

In the figures that follow, 1 mil means 0.001 inch.

A Probe Map Is Not Yet a Surface

The probe produces individual XYZ measurements. The cutter, however, moves continuously between those points. The software therefore has to estimate the surface height at every location along the engraving path. That process is called interpolation.

The challenge is to interpolate smoothly enough for machining without inventing geometry that the measurements do not support.

Start with the Simplest Reasonable Model

Our first model was a global quadratic surface. In practical terms, it is one smooth mathematical sheet that can represent overall height, tilt, bowl-like curvature, saddle-like curvature, and twist across the measured area.

That model is attractive because it is stable, predictable, and easy to evaluate. On a smoothly machined surface, it may be all that is needed.

The handworked production tooling exposed its limitation. One tailstock contained a measured point that differed from the best-fit quadratic by 0.00197 inch—about 20% of a 0.010-inch engraving depth. Across the production set, the largest departure from a best-fit quadratic reached 0.00360 inch.

The quadratic was describing the broad trend correctly, but it was smoothing over local geometry that mattered to the cutter.

Use a Smooth Local Model When the Data Requires It

For these surfaces, we moved to a thin-plate spline, or TPS. The useful mental picture is a flexible sheet pulled gently toward the measured points. Unlike one global bowl or saddle, it can respond to local changes. Unlike a faceted mesh, it remains smooth enough for a cutting tool to follow.

We also apply controlled smoothing. Forcing the surface through every measured tenth would make the cutter react to probe noise and insignificant measurement scatter. The goal is not to reproduce every fluctuation in the data. The goal is to recover the underlying metal surface accurately enough to control engraving depth.

Do Not Trust a Smooth Picture Just Because It Looks Convincing

A polished 3D plot can still be wrong. We therefore wrote custom software to challenge the machining surface in several independent ways before it is allowed to modify G-code.

First, the software builds a second surface from a Delaunay triangulation—a network of triangles connecting nearby measured points. Within each triangle, the surface is determined directly from its three measured corners. We generally do not machine from that model because the slope changes abruptly at triangle boundaries, but its local construction makes it a useful independent reference.

The software compares the smooth TPS surface with the triangulated surface beneath the actual engraving. Close agreement between two very different methods increases confidence. A meaningful disagreement is not resolved by choosing whichever plot looks better. It is resolved by adding another probe point in that region.

Second, the software performs a hidden-point test, commonly called leave-one-out validation. It removes one interior probe measurement, rebuilds the surface from the remaining points, and asks the model to predict the measurement it was not allowed to see.

This is more demanding than checking how closely a model fits the same points used to construct it. It tests whether the surface behaves sensibly between measurements.

On the representative tailstock, the global quadratic predicted the hidden points with an RMS error of 0.00103 inch. The smoothed TPS reduced that error to 0.00036 inch. On that real handworked surface, the local model predicted unseen measurements 2.9 times more accurately.

The same pattern appeared across the production set. Fourteen of 16 unique surface maps contained at least one local departure greater than 0.001 inch from the best-fit quadratic. Four exceeded 0.002 inch, and the largest reached 0.00360 inch. The TPS predicted held-out measurements more accurately than the quadratic on all 16 maps, with a median improvement of about 2.1 times.

The conclusion was not that a thin-plate spline should always be trusted. The conclusion was that no surface model should be trusted merely because it is smooth or mathematically convenient. The model had to earn its place in the machining workflow by agreeing with the measurements and predicting data it had not seen.

The Most Important Safety Rule: Do Not Extrapolate

Interpolation estimates the surface between measurements. Extrapolation predicts what happens beyond the measured region. That distinction matters enormously when the output will move a cutter against customer tooling.

The software draws the smallest boundary that encloses the outer probe points. Mathematically, this boundary is called the convex hull; visually, it is like stretching a rubber band around the outside measurements.

Every compensated cutting move must remain inside that boundary. Inside it, the software is estimating between observed points. Outside it, the software would be inventing a surface that was never measured.

If any required cutting point falls outside the measured boundary, the production workflow blocks G-code generation. The operator must expand the probe map. The software does not quietly continue into unknown geometry.

Warp the Toolpath, Not the Mold

Once the measured surface passes the checks, the custom software we wrote for this project processes the original CNC program.

The original CAM program already contains the machining intent: the RIC geometry, feeds and speeds, multiple depth passes, ramps, plunges, lead-ins, and lead-outs. We preserve that logic. The software changes only the Z position needed to keep each cutting move at the intended depth relative to the measured surface beneath it.

This is not one Z shift applied to the entire program. The correction changes continuously along the path.

Longer linear moves are divided into shorter segments so the tool can rise and fall smoothly with the surface. Cutting arcs are converted into short linear moves, each with its own local Z correction. A controlled extra depth can be added after test cuts, but it is applied only to cutting geometry. Retracts and clearance moves are not indiscriminately lowered.

The result is still the engraving created in CAM. It is simply no longer confined to a flat plane. It has been warped onto the physical production surface.

The software also makes the reasoning visible before the spindle runs. The operator can inspect the probe measurements, fit residuals, local-model comparison, hidden-point results, measured boundary, original flat path, and final compensated path. These checks do not eliminate machining risk. They expose assumptions and unsupported regions before those assumptions reach the tool.

From Compensated G-Code to Finished Tooling

The analysis mattered only if it produced a clean, durable mark on the actual production components.

Immediately after machining, the new RIC appeared as bright cut metal against the blended cavity surface. That contrast made the fresh toolpath easy to inspect, but it was not the final test.

After sandblasting, the cavity returned to a uniform matte finish. Any inconsistency in depth or line width would have been much harder to hide. The finished mark therefore provided a more demanding visual check than the freshly machined surface.

The finished tooling is the practical proof of the workflow. Fixturing, probing, surface reconstruction, model validation, G-code compensation, machining, inspection, and final finishing all had to work together on a high-mix family of components with materially different geometries.

A Practical Capability for Legacy Tooling

This project produced more than a one-off engraving program. It established a repeatable method and custom software for modifying a shallow feature on an existing surface when the current geometry is unavailable or unreliable.

The approach is particularly useful when:

• the surrounding mold geometry is already acceptable and should not be machined away;
• the new feature is small, shallow, and sensitive to local surface height;
• reliable CAD of the finished production surface does not exist;
• nominally similar tools have been individually repaired, polished, or hand finished; and
• the component can be fixtured and probed with adequate access around the machining area.

Potential applications include:

• resin identification codes and recycling marks;
• cavity numbers and traceability identifiers;
• date, production, or inspection markings;
• molded-in branding and customer-specific graphics; and
• other small engraved features on repaired, worn, or hand-finished tooling.

Targeted probing is not a replacement for full 3D scanning or complete reverse engineering. It is a fit-for-purpose method for localized, high-consequence machining where the cutter must follow the actual tool and creating a new reference surface would damage acceptable geometry.

Need to Update Existing Blow Mold Markings?

Updating a molded-in marking does not automatically require a new mold, a large re-machining operation, or a reliable CAD model of the current surface. Hybrid CNC Parts can evaluate whether an existing tool can be restored, mapped, and re-engraved while preserving the surrounding cavity geometry.

The most useful starting information is:

• a photograph of the existing mold surface;
• the approximate size and location of the feature to be changed;
• the number of cavities, inserts, or tailstocks involved;
• the tooling material;
• the existing marking; and
• the desired replacement marking.

The Core Lesson

When production tooling has a long physical history, the best surface model may be the surface already sitting in the machine.

Regulatory Note

Regulatory status reviewed September 4, 2026. This article describes a manufacturing capability and is not legal advice. See California SB 343 and CalRecycle’s Accurate Recycling Labels page.

With our expertise in complex machine programming, we can help you modernize and restore your mold tools, increasing their lifetime value for your organization.

Contact us today to discuss your mold tooling needs.