The few people that know the lab measurements, aren't at liberty to tell. So the only public information on stealth comes from modeling. However, stealth is complicated. It's not just specular reflections, which are easily modeled by physical optics, but a number of others:
As a result, the outcomes of modeling are all over the map. RAM is prominently excluded from most of these models, and some even miss radar blockers.
But combat experience allows us to make an estimate. It's well known that one F-117 has been shot down by a rather outdated radar-guided missile.
Some early media reports of the shootdown have speculated that the missile was "eyeballed", or that bomb bay doors were open. The former is impossible: the crew's located inside a windowless room, and the missile isn't on a turret. Open bay doors would explain detection, but not a shootdown, since these older SAM have separate, mechanically steered search and targeting radars, taking about a minute to launch. There's a Hungarian-written SAM simulator which shows the controls and operation of the S-125.
The reports available today show the SAM was operated in radar mode, as intended, tracked / lost track of the target 4 times, and ultimately one of the two launched missiles achieved lock and hit; the other missed. Here is the report from the battery's commander, translated from Serbian:
I arrived at the firing position around 20:30. There were no nearby targets in the air, but some were at greater distances in various azimuths. Suddenly, on the observation radar display, at an azimuth of 195, I spotted a target at a distance of 23 kilometers.
At a distance of 14-15 km and an azimuth of 210 degrees, the firing officer, Lieutenant Colonel Zoltan Dani, ordered to search for the target. The targeting radar's radiation was turned on. We radiated for more than 10 seconds unsuccessfully.
I saw the target again at an azimuth of 240 degrees and a distance of 14 km. The guidance officer's dials clicked, but the operators lost it. Just when I thought this attempt would also fail ... The dials clicked, and the operators locked on. Stable tracking, azimuth 242 degrees, distance 14.5 km.
The first missile launched, then the second after 5 seconds. Muminovic reports the first launched and locked, the second launched but not locked. (20:55, end of engagement)
This report is consistent with how the SAM operates. It shows that the S-125 was able to detect an approaching F-117A at 23 km and engage it at 14.5 km. That was its very limit, as indicated by spotty tracking. The aircraft was detected and lost again 3 times, until the final lock on and shot. The exact missile used was "5В27Д" aka "В-601Д". Its specifications are as follows: minimum target RCS - 0.3 m2, maximum range - 28 km.
From this, we can estimate that, even if the missile's "minimum RCS" is specified at maximum range, the F-117's RCS, with RAM, can't be lower than (14.5/28)^4*0.3=0.02 m2, and is likely closer to 0.3 m2. Or, in dBsm terms, between -16 and -3 dBsm. The former figure is optimistic, since minimum RCS is normally specified at the best-Pk range, which is usually around half of maximum range.
While the claims about -40 dBsm for the F-22 and -30 for the F-35 should be taken with caution (it's direct head-on at one frequency only), the consensus from most modeling is that it's no more than -16 dB without RAM, and likely -25 dBsm with RAM for the F-35.
The only other stealth aircraft with RCS without RAM in the -3 to -16 dB range is the Su-57, modeled here, which is consistent with the manufacturer's patent. And it's a sure bet that Lockheed's 3rd and 4th attempts at stealth are better at it than Sukhoi's first, especially being more of a "Silent Flanker" than a pure stealth design.
What has improved in technology since the F-117A includes:
- Composite skins. The F-117A is all-metal, and metal panels are susceptible to creep deformation. CRFP panels retain their initial shape much more accurately.
- Composite sandwich panels. While the main skin has to be a radar reflector to shield the internals, semi-transparent sandwich panels are very useful for control surfaces, which need to move.
- S-ducts that are heavy, but very effective at concealing the engines. The F-117 uses radar blocker grates to shield them, visible on the outside. Out of modern stealth aircraft, only the Su-57 uses that design.
- Thrust vectoring, allowing for quiet maneuvering without using the control surfaces, which make the shape less-stealthy while in use.
- Edge treatment. Semi-transparent fiberglass composites are used on the edges of modern stealth aircraft, greatly reducing traveling waves and edge diffraction.
- Data from the F-117A itself - lessons about what works for stealth and what doesn't, in a full-scale aircraft going through a brief but real lifecycle.
More recent reports also indicate that another F-117 has been detected and shot at by a similar SAM over Serbia, but only received minor damage. This article, coming from that plane's pilot, also details the lengths the US went to in order to keep the F-117s safe: avoiding "double-digit" SAM (SA-10 and above), providing F-16CJ SEAD escort with HARMs, and EF-111 and EA-6B providing radar-jamming.
Protecting a stealth aircraft with so many non-stealth ones isn't a sign of great confidence. But flying it in combat conditions was definitely worth it, for valuable lessons to improve future stealth designs.
Regarding composite vs metal, this is what the "flat" skin of the F-117A actually looks like. The wrinkles around the metal are pretty good at reflecting radar. This is further evidence that the F-117A is between 0.1 m2 and 1 m2 in terms of RCS. 0.3 m2 seems most likely, since that's what the S-125 is advertised at.
That said, the F-117A does have some great stealth features that have been lost in newer aircraft. One is the narrow slotted exhausts - they provide the lowest infrared and sound signatures of all shapes. Another is the first IRST on a US jet, which unlike even LPI radar is completely immune to interception.
Conclusion: the F-117 was an early attempt at stealth, made with limited knowledge, tools, and materials. Its shape reduces specular reflections, but these are just one of many kinds. It's likely that its designers have under-accounted for many variables.
Combat experience indicates the F-117's actual RCS to be significantly higher than predicted by a basic physical optics model. In particular, its S-band RCS is likely on the order of -3 dBsm, at best -10. Modern stealth design is considerably better in that regard, even for the Su-57, the least-stealthy of 5th gen fighters.