THE REAL STORY behind the four-year-old accident at the Three Mile Island nuclear plant finally seemed to start spilling out at the end of last year.

The press missed a lot of the story, and quite a story it was: evidence that operators at TMI falsified earlier reports on coolant leaks to the Nuclear Regulatory Commission; a study -- done by GPU itself more than a year before the accident -- which found that "the quality of operator personnel is on a continuous downward trend due to lack of training;" a suggestion of contradictory statements under oath by a key TMI employe; evidence of an operator cheating on a qualifying examination and much else.

The setting for these revelations was not one of the five official investigations held on the worst commercial nuclear accident in the nation's history. Those inquiries -- by two congressional committees, two special investigative panels and the Nuclear Regulatory Commission -- did not concentrate on drawing out such details, and did not fully develop the bits and pieces of evidence they had uncovered.

But now, in a federal courtroom in Manhattan, heaps of money were at stake, so no holds were barred. General Public Utilities, owner of TMI, was suing Babcock & Wilcox, maker of the TMI reactor, seeking $4 billion in damages, charging that B&W had caused "unthinkable risks to human life." Not standing still for that, Babcock & Wilcox retaliated with charges of "willful and wanton misconduct" by GPU at Three Mile Island, and the evidence about who was at fault began to flow.

It did not flow for long. In late January, just as Babcock & Wilcox was getting into its own attack during the trial, GPU announced that it was settling out of court for $37 million, and this in the form of rebates on equipment and services it would continue to buy over the next decade from Babcock & Wilcox.

But what emerged from the trial amounted to an invitation to a House Interior subcommittee -- scheduled to hold hearings Tuesday on the Nuclear Regulatory Commission's budget authorization -- to delve further. If it really wants to.

The subcommittee has asked the NRC to note any significant information on TMI that surfaced during the trial, as well as the status of the NRC position on a separate $4-billion suit GPU has filed against the government.

It's unclear, however, how much NRC officials will find it in their interest to volunteer on Tuesday. That second $4-billion GPU suit charges the NRC with negligence, claiming -- in a novel twist for the corporate world -- that the regulators didn't regulate firmly enough, that if the commission "had acted with due care... the TMI accident would not have occurred."

As one government official who has followed the TMI case says, "A lot of serious accusations have been made about the NRC's own competence in this case. Testimony in that [New York] trial showed that GPU falsified important data to the NRC. Although the [NRC] staff had plenty of opportunity to call the utility on that and other infractions, it didn't take the matter seriously. When the [NRC] case comes to court, logically somebody is going to want to know why."

But whether or not the NRC is interested in pursuing the leads that came out of the GPU-B&W litigation, that trial did produce some fascinating information. As Henry Myers, science adviser to the House Interior Committee, remarks of that aborted court battle: "For the first time there were people who were interested in the unvarnished truth, and they were determined to root it out."

GPU may not have been eager to file suit against B&W or the NRC. But lawsuits were needed to satisfy two goups that hold great power over the utility: stockholders and state public utility commissions.

Both must be convinced the utility did all it could to seek outside funds to pay the nearly $1 billion in TMI cleanup costs and the $3-billion expense of replacement electrical power while TMI is out of service.

Despite dogged GPU efforts to obtain federal funds and contributions from other nuclear users, there appear to be only two sources of money -- GPU's investors and customers. Investors aren't standing in line to buy the company's stock. That leaves GPU's ratepayers in central Pennsylvania and New Jersey, who will continue to be assessed for a substantial share of the cleanup costs. If GPU hopes to win more rate increases, it must demonstrate to the utility commissions that it made good-faith efforts to obtain funds from other sources.

So GPU found itself in the courtroom of Federal District Judge Richard Owen on Nov. 1. David Klingsberg, a noted New York lawyer retained by GPU, opened the trial by promising to relate "a frightening episode of corporate misconduct" by Babcock & Wilcox.

The keystones of GPU's negligence claim were two memos written in February 1979 by a B&W engineer describing a September 1977 incident at an Ohio nuclear plant. The Ohio problem, also involving a loss of coolant to a B&W-built reactor, was indentified and corrected in 20 minutes. The memo advised B&W management to warn other BPW reactor owners to revise emergency instructions to prevent a recurrence. (The warning finally was sent out after the TMI accident.) The memos, Klingsberg declared, "are the smoking guns of this case."

B&W's attorney was Robert B. Fiske Jr., former U.S. attorney for the Southern District of New York, who responded forcefully.

"We expect to prove that GPU was guilty not merely of ordinary negligence, but of recklessness and in fact engaged in certain respects in deliberate, willful and wanton misconduct," he declared. When problems began in the reactor, he said, operators had emergency instructions available but ignored them in favor of their own "on the spot, off the cuff, ad hoc analysis of system conditions, for which... their training program left them totally ill-equipped..."

It was during cross-examination of GPU's own witnesses that Fiske built much of his case. The cross-examination of William Zewe, shift supervisor in TMI Unit 2 at the time of the accident, was an example.

At issue was the temperature of steam surging out of a stuck-open valve on the reactor's pressurizer tank. TMI operators didn't know for two hours and 20 minutes that the valve was stuck open, allowing much of the reactor cooling water to pour out. They compounded the problem by shutting off emergency make-up water.

GPU contended that faulty B&W instructions caused operators to misread the problem. But B&W argued that any competent operator should have been tipped off by the abnormally high temperatures, caused by the escaping steam, that the valve was stuck open. To prove it was incompetence -- not B&W's instructions -- at fault, Fiske had to show that Zewe knew of the high temperatures for a long time and did not respond.

To begin with, Zewe told Fiske that he hadn't followed GPU's booklet of operating procedures for abnormal situations as the accident unfolded. He also said that he had been told by a control-room colleague that temperatures had dropped from a high of 280 degrees Fahrenheit to 230 degrees, indicating to him the valve had closed.

That version, Fiske said, was different from previous sworn testimony by Zewe before other investigatory bodies. On one occasion, Fisk said, Zewe had testified that he had heard the temperature incorrectly. Fiske also recited earlier Zewe testimony before a Senate committee probing TMI. When asked what he considered the "major contributor to this accident," Fiske said, Zewe had responded then: "Failure of the operator, myself, to adequately determine the electromagnetic relief valve was still open."

Fiske accused Zewe of altering his testimony now to offset any inference of incompetence on his part. Zewe denied the charge, saying that his testimony on this occasion was to the best of his recollection.

The liveliest evidence about conditions at TMI came during the cross-examination of Robert C. Arnold, president of GPU Nuclear Corp. and, as such, TMI's chief overseer. In Arnold's view, maintenance of the TMI reactor was "adequate." Operator training was "acceptable." Employes responsible for running the reactor were reliable and capable.

But for every positive assessment by Arnold, Fiske countered with other evidence -- all drawn from GPU's own files, from its employes' sworn statements during previous investigations or the trial itself, and from NRC documents. This evidence included:

* The TMI station manager advised his superiors before the accident that poor maintenance at the plant was a serious problem. At the time of the accident 800 to 1,000 repair tasks were pending. Arnold estimated under cross-examination that TMI's maintenance budget for 1979 had been cut "several hundred thousand, and [it] could easily have approached a million [dollars]," but he contended that reactor maintenance was still sufficient.

* The TMI reactor was fraught with problems from the beginning. Starting in 1979, Unit 2's coolant system leaked at rates that frequently exceeded allowable NRC limits. NRC rules required that the reactor be shut down to determine the cause of such leaks. But Fiske cited NRC findings that TMI operators, to avoid the expense of shutdowns, manipulated the leak rates, falsely certifying to the NRC that there was no major problem. When the NRC learned of the falsified records after the TMI accident, Fiske said, it fined GPU for the infraction.

* A GPU audit conducted more than a year before the accident noted that "the quality of operating personnel is on a continuous downward trend due to lack of training." Attendance at operator training classes, to prepare for NRC-required exams, dropped as low as 33 percent. The TMI supervisor of training was devoting half his time to studying for his own operator's license, the GPU audit showed, and when he failed the test, he spent all his time preparing for a re-examination.

* A control-room supervisor, who had to be recertified to the NRC following the accident, was found to have cheated on the exam. He submitted his test with answers in someone else's handwriting and a different color ink.

Despite all this, Arnold defended the training program and took exception to Fiske's statement that TMI management "failed to instill a sense of respect for the training" of its operators.

This last assertion by Arnold elicited some sharp remarks from Judge Owen, who said he had gotten the impression from Arnold's direct testimony that "everything [at TMI] was hunky-dory." In fact, said the judge, "the operator training here was pretty atrocious.... That bears on the quality of operator skill, which bears upon the question of whether or not they should have recognized certain symptoms and closed the relief valve early on in the game."

On Jan. 24, with the defense still in the early stage of calling its own witnesses, GPU suddenly agreed to settle. The full text of the settlement agreement has not been made public, but one portion that was released states that "neither party has established that the other was the cause of the TMI-2 accident."

By most assessments, the settlement was forced upon a beleagured GPU management by its increasingly worried colleagues in the nuclear industry. "This had the potential for dragging down an entire industry that's already on its knees," commented a government lawyer following the case. "Nobody in all the TMI investigations had brought out such damaging information."

The question now is whether anybody in Congress or elsewhere really has an interest in pursuing the rest of the story, in bringing out that full "unvarished truth" about what really happened at Three Mile Island.