Close the loop
Check that the spec, the program and production still agree — and when they do not, see who owns the number. The dashboard joins every requirement to its test and its production data, reconciles the spec's limits against the program's and the program's against what the latest lot actually ran, and attributes every disagreement to the artefact that owns it.
One surface where the spec, the program, the load board and production data meet. It recomputes its verdicts from the live artefacts on every load and labels which copy of the program it read — the open workbook or the last ingested one.
A requirement no test implements, and a program test no requirement asks for, are both listed.
Program limits and force conditions are reconciled against the spec. A program looser than the spec is a compliance error.
The limits the latest lot recorded are compared with the program as it stands now, and a difference is reported as drift.
Program pins are checked against the board's nets, and each requirement is followed through to routed copper.
Every verdict comes from a deterministic engine: there is no language model in the pipeline, so the same artefacts always give the same answer. Ask the IG-XL agent whether the spec's guard bands reached the program and production and it calls the same reconcile, returning the spec, program and production limits side by side with the mismatch named.
Test names are allowed to drift between spec, program and lots, because those joins key on test number. Board joins key on net names and allow only known DUT/DIB prefixes and tester-side suffixes; a net renamed any other way is reported as unmatched, never guessed.
| Join | What it catches |
|---|---|
| Spec ↔ Program | Coverage gaps (a requirement no test implements) and limit or force-condition drift — a program limit looser than the spec is a compliance error, attributed to the pillar that owns each number. |
| Program ↔ Board | A program that drives a pin the board does not carry, a board pin no test uses, a power rail mapped to a digital channel. |
| Schematic ↔ Copper | A net the schematic declares but no routed copper carries — so a requirement that looks covered on paper is exposed as unreachable on the physical board, tester-side series-termination and Kelvin nets included, disclosed as such. |
| Program ↔ Production | The limits a lot actually ran versus the limits the program has now — historical drift is reported as drift, never mislabelled as a spec violation. |
Every load reads the four pillars fresh — the ingested spec, the program workbook, the saved board design and the production STDF — and compares values by what they mean, so a limit written in milliamps in one artefact and in amps, in scientific notation, in another compares as equal.
Attribution is the point. In a real project the limit sources legitimately differ, and the difference between “the lot ran at a recorded limit” and “the program violates the spec” is the difference between a finding and a false alarm. The joins compose into one row per test — spec → program → nets → copper → lot Cpk — and a marginal test's row lists the board components most likely involved. A missing pillar renders as awaiting data with the action that unlocks it, never as a silent pass.
The check starts earlier, too. Right after generation, the program's limits are checked against the spec, and a program looser than the spec gets a named verdict on its run page.
The chain does not stop at the netlist. Select a requirement in the design's layout tab and its nets — including the tester-side series-termination and Kelvin force/sense extensions, disclosed as such — light up on the routed board as the layout tool drew it. The copper hop reads the board bound to the project, imported from the EDA tool that drew it.
On the demonstration project — a program generated for a small motion sensor with a 4-wire SPI interface, with demonstration lots of seeded synthetic data — spec, program and production join on test number for all 16 requirements. The limit reconcile reports one compliance error and no drift warnings.
The error is a guard band planted in the demonstration data: the spec's upper limit on the power-down supply-current test was tightened after the program was generated, the program and every lot kept the old value, and the reconcile names it “program looser than spec”. Asked whether the spec's guard bands had reached the program and production, the IG-XL agent returned the spec, program and production limits and named the same mismatch; the three values match the database.
Measured on the demonstration project: presence on test number, not correctness. Lots are seeded synthetic data with planted defects. The copper hop is not shown there: no routed vendor board is bound to it yet, and the dashboard says so rather than drawing one.