186: Chapter 186 8.5x Dimensional Reduction Crushing

October 13th, 7:58 AM.

University of Science and Technology of China, Quantum Laboratory.

The four main screens in the center of the control room had all switched to the global monitoring page for the joint experiment.

On the far left was an extremely complex chip physical adjacency map.

The second screen showed the readout electronics groupings, arranged like a spiderweb.

The third screen displayed a nanosecond-precise timeline of control instructions.

And on the rightmost screen hung the 186 stable correlation edges that Jiang Lin had frozen yesterday.

They were no longer a tangled mess; instead, they were strictly categorized according to the four classes established by Jiang Lin, differentiated using striking colors.

[Red] Seventy-nine readout correlation edges.

[Blue] Forty-eight control-loop correlation edges.

[Gray] Forty-one undetermined-source edges.

[Green] Eighteen Physics correlation candidates.

The laboratory project leader stood behind the control console with a solemn expression.

The three core research teams responsible for readout, hardware control, and the decoder were seated at their respective terminals, hands hovering over their keyboards.

Jiang Lin was connected via a controlled, encrypted video link.

Exactly 8:00 AM.

"First round of readout intervention, begin."

Following the project leader's command, the operator hit Enter.

There were no changes to the Physics connections at the hardware layer.

The quantum chip inside the dilution refrigerator, the resonant cavities, and the spatial adjacency relationships between the Physics qubits all remained completely untouched.

What the experimental team actually changed was the acquisition path.

They swapped the digital demodulation chains and clock groupings for two compatible readout channels.

The weak readout signals originally processed by the first digital link were forcibly rerouted to the second.

The synchronized clocks, demodulation parameters, and data buffering paths that originally belonged to the second group were swapped to the first.

This was a direct cross-validation.

If a correlation edge truly originated from the Physics structure on the chip—such as local crosstalk or non-computational state leakage—then no matter which path the backend digital demodulation used for observation, it should not jump from one pair of qubits to another Physics region.

But if it originated from the acquisition system, it might migrate along with the digital link.

The first batch of data began rolling back twenty-one minutes later.

Among the seventy-nine red readout correlation edges, the top-ranked one was numbered E-031.

In the original old data, it bizarrely connected two detection nodes that were physically far apart.

The only explicit commonality between these two nodes was that they had long shared the same demodulation clock group.

As the readout grouping swap completed, results appeared quickly.

The correlation strength between the original two nodes rapidly dropped from 0.362 to 0.044, falling back into the background noise range.

Meanwhile, between the other two nodes newly connected to this digital link, a new correlation edge with a strength of 0.339 appeared.

Its Physics location had changed, but the delay, frequency distribution, and window performance of the correlation structure were almost identical to the previous E-031.

It had followed the readout link.

The researcher responsible for readout swallowed hard and overlaid the topology maps from both experiments, then typed furiously to bring up real-time tracking for more candidate edges.

On the screen, new results were continuously written into the classification table.

9:14 AM.

The first round of readout intervention was completed.

[Control Intervention Cumulative Review Report]

[Cumulative Included in Review: 48, including E-009 attributed yesterday]

[Confirmed Disappeared or Migrated with Control Action: 45]

[Reclassified to Undetermined Source: 3]

Seventy-five edges obtained direct intervention evidence.

Most notable among them were seven edges that did not migrate with the groupings, but instead returned directly to the background range after recalibrating the fixed phase deviation between the two clock groups.

The experimental team traced these seven edges back through the codebase, eventually pinpointing a synchronization compensation parameter that had been in use for three major versions.

Its original design intent was to offset the fixed Physics delay between the two readout links. However, after the fourth major hardware upgrade, the legacy parameter was not updated along with the new data acquisition cards.

Instead of eliminating delay, this expired parameter consistently created cross-round time offsets during each measurement, thereby generating seven stable correlation edges.

The researcher in charge of readout stared at the commit date in the Git version history, rendered speechless for a long moment.

This parameter had been viewed by the team as merely an engineering detail prior to data cleaning.

In the correlation analyses over the past few months, no one had considered it part of the noise generation mechanism. Yet it had quietly persisted behind seven stable correlation edges across three versions.

The project leader broke the silence: "Should we run another round after fixing this parameter?"

"It's already running," the researcher replied, voice slightly raspy.

Eleven minutes later, all seven edges receded into the meaningless background range.

Without hesitation, the readout team wrote the fix into the global engineering log in bright red font.

[Digital readout clock compensation parameters must not be inherited independently of hardware versions.]

Watching all this from the other side of the screen, Jiang Lin had no intention of performing any further decoder optimization on these seven edges.

Problems that could be eliminated at the source in the acquisition system did not need to be adapted to by the decoder.

9:42 AM.

"Control intervention begin."

The laboratory retained the current decoder's judgment on error locations and only changed how correction actions were issued.

Interventions such as delay, reordering, bypassing, and judgment-only without dispatch were executed on the four categories of high-frequency local correction paths.

Attribution for E-009 had been completed yesterday. What needed to be re-examined today were the remaining forty-seven control-loop correlation edges.

An hour later, the first round of control intervention ended.

Out of the forty-seven edges, thirty-one weakened rapidly as their corresponding control paths were disabled.

Nine migrated along with the new dispatch locations.

For four of them, after the correction actions were delayed, their correlation characteristics shifted from spanning one round to spanning two rounds.

Only three showed no significant change in place after the control actions were modified.

The researcher in charge of control pulled one of them onto the main screen.

[File: E-118]

[Original Classification: Control Loop Correlated]

[Post-Intervention Change: No significant change in Physics location or strength]

This edge was located between two adjacent detection nodes.

In previous historical data, it highly overlapped on the timeline with a certain category of local correction actions, so it had been temporarily categorized under the control loop directory.

But after halting that action, its correlation still persisted.

The PhD student responsible for the decoder poked his head over and asked Jiang Lin, "Does that mean the control action just happened to occur simultaneously with it, and wasn't actually the cause triggering it?"

"Possible," Jiang Lin answered bluntly. "Move it to the [Undetermined Source] directory. The first round of classification was just a working hypothesis; if the intervention results don't support it, it must be reclassified."

Under Jiang Lin's command, the color of E-118 on the screen changed from blue to gray.

The other two edges that likewise showed no significant change were also transferred to Undetermined Source.

11:37 AM.

The control intervention results were officially frozen.

[Control Intervention Review Report]

[Original Total Entering Review: 48]

[Disappeared or Migrated with Control Action: 45]

[Reclassified to Undetermined Source: 3]

By this point, clear intervention evidence had been gathered for the vast majority of the seventy-nine readout correlation edges and forty-eight control correlation edges.

Now, everyone's attention shifted to the smallest group.

[Eighteen Physics Correlation Candidates]

12:05 PM.

The laboratory completed a full recalibration.

The readout groupings maintained the newly verified mappings.

The control system adhered to the cleaner post-intervention dispatch strategy.

If any structures among the eighteen edges were merely remnants left behind by the old readout system or control strategy, they ought to disappear during this system-level recalibration.

The only thing that remained unchanged was the Physics adjacency layout on the chip.

1:26 PM.

The first batch of data for the new window flooded into the review terminals.

The eighteen Physics candidates were repositioned one by one.

2:48 PM.

The eighteen edges completed their first round of elimination.

[First Elimination Report for Physics Candidates]

[Retained Original Physics Location and Temporal Structure: 11]

[Disappeared with Readout or Control Changes, Transferred to Corresponding System Directories: 4]

[Insufficient Evidence, Transferred to Undetermined Source: 3]

In the end, only eleven remained.

These eleven edges did not migrate with the digital demodulation chains, did not change with control strategies, nor did they disappear due to clock alignment fixes.

After system recalibration, they remained locked onto their original Physics adjacency locations.

This still did not prove which specific microscopic mechanism they corresponded to; it only showed that within the scope of current interventions, no evidence was found that they were generated by the readout or control systems.

They remained preserved as Physics correlation candidates.

Looking at the meager number, the project leader asked Jiang Lin, "All eleven of these—put them straight into the decoder?"

"Not yet," Jiang Lin remained cautious.

He pulled up the program and re-clustered these eleven edges according to their Physics locations.

Seven of them were concentrated in two adjacent regions.

They all overlapped with measurement rounds containing two-qubit gates and exhibited a noticeable increase in the subsequent round.

The remaining four were relatively scattered, sharing neither the same gate operation nor the same delay structure.

Staring at the seven clustered edges, Jiang Lin suddenly asked, "In your low-level logs, do you have independent monitoring metrics related to non-computational state leakage?"

The hardware researcher froze for a moment, flicked through records, and replied, "There is a set of auxiliary criteria, but it wasn't included in the raw data packet; it's not an official error syndrome field."

"Pull it up immediately."

Five minutes later, the heatmap of the seven clustered correlation edges and the auxiliary criterion curve for non-computational state leakage were placed on the same timeline.

The peaks coincided almost perfectly.

When local leakage criteria surged, the probability of occurrence for these seven edges increased synchronously.

When the leakage criteria dropped, they receded accordingly.

The hardware researcher took a deep breath: "But that only proves they vary together; it doesn't directly prove leakage is the root cause."

"Then we intervene," Jiang Lin said.

The experimental team alternately ran control batches and leakage-clearing batches in the same target region, keeping cycle duration and other experimental conditions identical for both groups.

3:39 PM.

The control batch finished running.

All seven edges remained in their original locations, with an average correlation strength of 0.286.

4:16 PM.

The intervention batch finished.

The peak of the non-computational state leakage auxiliary criteria dropped significantly.

Among the seven correlation edges, six saw their correlation strengths synchronously drop by over sixty percent.

The last one dropped by less than twenty percent and still required separate analysis.

The project leader reviewed the set of intervention results twice: "So can these six be officially categorized under [Leakage-Correlated Physics Errors]?"

"We can upgrade them to [High-Confidence Physics Candidates]," Jiang Lin corrected. "But we cannot claim we've reconstructed the complete microscopic mechanism. Current evidence can only take us this far."

"And what about the seventh one that didn't drop much?"

"Keep it separate, retaining its candidate status."

Before entering the decoder, the eleven edges were further split into two categories.

[Leakage-Induced Physics Candidates: 6 edges]

[Stable Local Physics Correlation Candidates: 5 edges]

Wherever the current data could provide support, the classification stopped at that exact step.

The PhD student in charge of the decoder regenerated the final weighted matching graph on the terminal.

Previously, this graph contained sixty-four seemingly stable edges.

Now, with seven candidate edges removed, only eleven remained in the decoder.

5:08 PM.

System frozen.

[Decoder Deployment Protocol]

[Version ID: Conservative_Correlation_Decoder_v1.0 (Conservative Correlation Decoder v1.0)]

[Selected Physics Correlation Edges: 11]

[Readout Correlation Edges: Forbidden]

[Control Loop Correlation Edges: Forbidden]

[Undetermined Source Edges: Forbidden]

[Code Hash: 7f2ba1 (Generated)]

[Status: Current round candidate edge list frozen.]

Looking at the list, the project leader tentatively asked, "Jiang Lin, should we run a test with this afternoon's data first to see the performance?"

"No," Jiang Lin refused.

"Why? The model is already frozen, isn't it?"

"Because the data generated this afternoon participated in candidate edge classification and intervention decisions," Jiang Lin explained. "To prove this method works, final performance must be evaluated in a new window that no one—including me—has ever seen."

The laboratory immediately initiated a second full calibration.

Once the new window was established, all parameters were remeasured.

Jiang Lin did not touch the low-level data, nor was the frozen eleven-edge decoder allowed any further modifications.

7:43 PM.

A brand-new independent calibration window was established.

[Blind_Window_02]

This blind test was divided into four groups.

[Group A: Original readout and control configuration + Independent noise baseline model]

[Group B: Original readout and control configuration + 64-edge correlation model]

[Group C: Post-fix readout and control configuration + Independent noise baseline model]

[Group D: Post-fix readout and control configuration + 11-edge conservative decoder]

The original readout/control configuration and the post-fix configuration ran alternately according to pre-frozen random batches.

The raw error syndrome streams generated by the two configurations were submitted to their respective frozen decoders for recalculation.

Ultimately yielding four sets of results: A, B, C, and D.

Estimated runtime: four hours.

Jiang Lin exited the remote video window.

From this moment on, any added ad-hoc explanation or threshold modification due to unpromising data would render the blind test meaningless.

11:58 PM.

Room 402, Zijing Building No. 17.

The fixed communication window of the Research Support Unit lit up.

Professor Lu Zhixing sent an encrypted results file with a double digital signature.

Jiang Lin checked the hash; it matched.

Jiang Lin double-clicked to open the file.

The file contained no analytical commentary, only four lines of numbers.

[Blind_Window_02 Blind Test Results]

[Group A: 1.000]

[Group B: 1.074]

[Group C: 0.936]

[Group D: 0.898]

The results were clear enough.

In a completely unfamiliar calibration window, the 64-edge model degraded the proxy logic error rate by 7.4%.

Group C proved that simply fixing the readout clock deviation and severing the erroneous control loop reduced the proxy logic error rate by 6.4%, even without incorporating any Physics correlation edges.

Building on this, Group D incorporated eleven intervention-verified Physics correlation candidates into the decoder, driving the result down further to 0.898.

Relative to the post-fix baseline, the eleven Physics candidates provided an additional improvement of approximately 4.1%.

Compared to Group A, the proxy logic error rate saw an overall reduction of 10.2%.

[part:gemini-3.5-flash-lite]

This result proves that the causal identifiability review protocol and eleven-edge decoder established by Jiang Lin can maintain gains across new calibration windows.

October 14th, 8:12 AM.

The controlled video conference was re-established.

The project leader, hardware team leader, control team leader, and the PhD student in charge of the decoder from the USTC Quantum Laboratory were all online.

Jiang Lin wrote last night's result of a 10.2% drop into the project homepage, but did not close the project.

The project leader looked at the status bar.

[Syndrome_Correlation_Ledger / FINAL_SCALING_TEST_PENDING]

"The code distance scaling test is still pending." The project leader said in a deep voice.

"Right." Jiang Lin said.

"What kind of industry data are you planning to use as a reference?"

Before Jiang Lin could speak, the PhD student in charge of the decoder had already pulled up a preprint published this July from the literature database.

The authors' affiliation of the paper was displayed at the top of the screen.

[Google Quantum AI]

Below were two sets of experimental data that had been repeatedly discussed in the surface code field for months.

[Google Public Reference Data]

[Code Distance 3 Average Logical Error Rate Per Round: 3.028%]

[Code Distance 5 Logical Error Rate Per Round: 2.914%]

[Relative Error Suppression After Scaling Code Distance: Approx. 3.8%]

Increasing the code distance of a surface code essentially uses more Physics qubits to protect one logical qubit.

If the larger the code distance, the higher the logical error rate instead, this error correction architecture could not be further scaled.

The results published by Google in July reduced the logical error rate per round from 3.028% to 2.914%, a relative decrease of about 3.8%.

Although the reduction was not large, it was the first time it was proved in experiments that after increasing the scale of the surface code, the error correction gains could exceed the extra errors brought by the newly added Physics qubits.

The project leader closed the literature page and opened the code distance scaling records accumulated by the USTC team over the past eleven months.

[USTC Historical Test Records]

[Code Distance 3 Average Logical Error Rate: 2.86% / round]

[Code Distance 5 Average Logical Error Rate: 3.11% / round]

[Change After Scaling Code Distance: 8.7% Deterioration]

If looking at absolute numbers alone, the code distance 3 system of USTC was 2.86%, lower than Google's public 3.028%.

But absolute numerical values between different devices cannot be directly ranked.

However, when the system was expanded from code distance 3 to code distance 5, the error rate did not decrease, but instead rose to 3.11%.

The hardware team leader looked at those two rows of numbers: "Jiang Lin, at least from the existing tests, our code distance 3 performance is not bad. The problem lies after the system is scaled up."

The PhD student in charge of the decoder chimed in: "At code distance 3, the system is relatively simple, and many hidden readout synchronization and closed-loop control problems have not yet formed a stable structure. After scaling to code distance 5, the lines increase, the control closed-loop becomes complex, and the footprints stepped on by those systems appear intensively, misleading the decoder."

The project leader asked the PhD student: "In these eleven months, what was the data of the best scaling test we ran?"

The PhD student smiled wryly and pulled up a row of records.

[Historical Best: Code Distance 3 2.82%, Code Distance 5 2.97%, still deteriorated by 5.3%.]

A full twenty-seven complete calibration windows, not once could the error rate be pushed below the code distance 3 baseline after scaling the code distance.

This meant that although USTC could build a code distance 5 surface code, continuously extract error syndrome data, and run through the complete decoding process, it could never prove that investing more Physics resources could bring fewer errors.

The project leader looked at Jiang Lin: "If this time we use the same identifiability review and conservative decoding scheme, what if code distance 5 still cannot be lower than code distance 3?"

"Then fall back to the origin and continue to look for the cause." Jiang Lin's answer did not have the slightest hesitation.

"Then this third-party support project will not be closed?"

"Not closed."

Jiang Lin looked at everyone on the screen.

"Distinguishing the sources of more than 180 relevant edges can only prove that the attribution method holds. Only when code distance 5 is stably lower than code distance 3 can this system be considered to have evidence for continued scaling."

9:04 AM.

The code distance scaling review plan was officially frozen.

The two sets of surface codes used the same twenty-five rounds of logical storage tasks, the same loop definitions, and the same evaluation standards.

It needs to be additionally explained that those eleven edges belonged to the Physics topology of code distance 3 and could not be moved to code distance 5 as they were.

In the previously reserved code distance 5 development window, USTC completed independent screening according to the same set of identifiability protocols, generating a conservative correlation list corresponding to the code distance 5 topology.

This list did not touch the next three official windows and was frozen before the experiment started.

Both code distance 3 and code distance 5 used the repaired readout clock configuration, adjusted control strategies, and conservative decoders frozen under the same protocol respectively.

The formal experiment spanned three separately recalibrated windows.

After each window ended, the results were sealed immediately. Jiang Lin could not touch any intermediate data.

October 14th, the first round started.

Now, the only thing truly placed in front of everyone was two sets of numbers.

Code distance 3.

Code distance 5.

October 16th, 5:36 PM.

The test process of the third window finally ended.

The Research Support Unit began to verify the freezing conditions of the three windows.

Code hash values, the three comparisons were completely identical.

Decoder version list, the three comparisons were completely identical.

The three windows completed independent calibration respectively, without sharing fitting parameters.

After confirming that everything was correct, the USTC team sent the summary results to Professor Lu Zhixing.

Professor Lu Zhixing completed the secondary digital signature confirmation.

6:12 PM.

The controlled video link was re-established.

The encrypted result file secondary-signed by Professor Lu Zhixing entered the USTC Quantum Laboratory and Jiang Lin's fixed receiving window simultaneously.

Both sides checked the hashes, and they matched.

The project leader entered the unsealing key.

The results appeared on the screens of both places simultaneously.

Jiang Lin double-clicked to open the file.

The first page was the result of the first calibration window.

[Window_01]

[Code Distance 3: 2.84% / round]

[Code Distance 5: 1.96% / round]

The second page, the second calibration window.

[Window_02]

[Code Distance 3: 2.88% / round]

[Code Distance 5: 1.92% / round]

The third page.

[Window_03]

[Code Distance 3: 2.86% / round]

[Code Distance 5: 1.94% / round]

Jiang Lin scrolled the mouse and came to the last page.

Summary of the three recalibration windows.

[Final Comprehensive Evaluation Report]

[Code Distance 3 Average Error Rate: 2.86% / round]

[Code Distance 5 Average Error Rate: 1.94% / round]

[Relative Error Suppression Magnitude After Scaling Code Distance: 32.2%]

[Separately Recalibrated Windows: 3]

[Same-Direction Results: 3 / 3]

Jiang Lin leaned back in his chair, pulled up a script, and placed USTC's results side by side with Google's public data from July on the screen.

[Google Quantum AI Public Results]

[Code Distance 3: 3.028%]

[Code Distance 5: 2.914%]

[Relative Error Suppression Magnitude: 3.8%]

[USTC Current Joint Review Results]

[Code Distance 3: 2.86%]

[Code Distance 5: 1.94%]

[Relative Error Suppression Magnitude: 32.2%]

Three point eight percent.

Thirty-two point two percent.

Calculated based on this relative suppression value, USTC's current result was about 8.47 times that of the public reference.

As the data transmission of the last independent calibration window was completed, in the USTC quantum control room, the PhD student in charge of the decoder swallowed a mouthful of saliva, his Adam's apple rolling with difficulty.

With slightly trembling hands, he typed the final summary enter key on the terminal.

On the main screen, the data stopped jumping and froze as the final result — [Relative Error Suppression Magnitude After Scaling Code Distance: 32.2%]

Not the 3.8% publicly released by Google.

It was 32.2%.

Nearly eight and a half times dimensionality reduction crushing!

"Clatter."

It was unknown whose carbon pen dropped to the ground.

The project leader felt a wave of intense dizziness.

He recalled the next-generation chip layout redo plan he had personally signed a few days ago.

If it weren't for Jiang Lin's swift and decisive dismantling of the entire causal chain these past few days, their entire team would have spent a round of tape-out, packaging, cryogenic calibration, and half a year of R&D time of the hardware team to patch a ghost vulnerability that didn't exist at all.

Cold sweat instantly soaked his back, but following closely behind was an ecstasy that rushed straight to the crown of his head.

"Verify the raw data."

As soon as he spoke, he found that his voice was extremely hoarse, carrying an uncontrollable tremor.

"Quickly, three comparisons, is there any abnormal spike?"

"None!" The PhD student's eye sockets were red, and he abruptly stood up, his voice nearly cracking with excitement. "All three windows separated smoothly without any abnormalities. The data is completely valid. We, we've pushed the error rate down by thirty percent."

The entire control room was like an active volcano about to erupt, suppressing earth-shattering fanaticism.

Everyone's gaze was glued to that large 32.2%, breathing heavily.

And on the other end of the video conference.

Building 17 of Zijing, Room 402.

Jiang Lin looked at the result on the screen—which, once passing external independent review, would be enough to attract high attention in the global quantum error correction field—with no ripples on his face.

He merely clicked open the confidence intervals, task definitions, and frozen hashes of the three windows one by one.

Seeing all consistent results, he asked calmly: "Have all three verification windows been reviewed?"

"All three internal verifications are completely consistent, and no statistical anomalies have been found so far. The raw data, code, and calibration logs have been frozen, and it is fully ready for external review." The project leader answered almost like standing at attention.

"Alright." Jiang Lin neatly pulled up the acceptance form. "Then submit the project closure, and I'll close the window on my side."

"Wait a minute." The project leader stopped him, and then deleted [External Method Support] from the independent review application.

Re-entering —

[Proposer of Core Attribution Method and Verification Protocol: Jiang Lin]

[Achievement Ownership: Listed separately, not incorporated into the collective contribution of the experimental team]

[Subsequent Papers, Achievement Registration, and Award Applications: Confirmed with priority according to actual contribution]

The project leader completed the electronic signature: "We will handle the subsequent reviews and materials ourselves, and won't take up your time anymore. Of course, the part of the contribution belonging to you will be clearly written in the official documents by USTC."

Jiang Lin glanced at it and said indifferently: "The top priority is to finish the review first."

"Of course."

"Alright, then that's it. You guys stay busy."

Then all the researchers present saw the video screen on Jiang Lin's side go black.

However, after the video screen was cut off, the remaining warmth of fanaticism in the control room lingered for a long time.

But the project leader knew that now was not the time to indulge emotions.

He took a deep breath, forcibly suppressed his thumping heartbeat, and turned back to the main control console.

"Since the three windows have passed internal cross-verification, and E-009 has also been confirmed to be induced by the control strategy, all hardware schedules based on the old conclusions must be re-examined."

He strode to the terminal and pulled up the engineering task schedule of the next version of the experimental chip.

At the very top of the schedule, a high-priority task that had been promoted for more than two months was prominently listed.

[Engineering Codename: E-009 Related Region Coupling Structure Redo]

[Status: Urgently in progress, waiting for final freeze at the end of the month.]

[Resource Estimation: Involves layout redrawing, a new tape-out cycle, and packaging calibration, estimated to take more than half a year.]

If it were before today, this task would have been regarded by the entire team as a life-saving straw to save the error correction rate.

But looking at it now, it was nothing short of an absurd joke.

The hardware team leader stepped forward and issued an instruction without any hesitation: "Cancel, cancel the whole case."

As permissions were confirmed, dazzling red characters covered the screen.

[Status Changed To: Permanently Cancelled]

[Cancellation Reason: The originally highly trusted Physics correlation has been confirmed to be a closed-loop control induced structure.]

Looking at the crossed-out task line on the screen, the project leader let out a long breath of foul air.

This hardware revision personally signed and promoted by him had almost buried half a year of hard work of the entire laboratory.

If they had really foolishly gone to rebuild the chip, even if the new version of the chip temporarily no longer had E-009, the real control inducement would still not have been eliminated.

After the next policy update, new pseudo-correlations might still appear.

Because they had found the wrong root cause from the very beginning.

Without Jiang Lin's dimensionality-reduction-strike-like review protocol, they would still be spinning around in endless failures to this day.

"Write this process of erroneous decision-making faithfully into the project review." The project leader looked around, his voice steady and powerful. "No deletions, no downplaying. I will take full responsibility."

No one made a sound at the scene, but the drive in everyone's eyes was almost overflowing.

After clearing away the historical baggage, the project leader clicked on another core task that had been backlogged for three months.

[Strategic Task: Dual Logical Physics Qubit Gate High-Fidelity Experiment]

[Current Status: Paused]

[Pause Reason: The effectiveness of code distance scaling has not been confirmed, and the single logical qubit protection capability is not enough to support the noise superposition of dual qubit gates.]

He gripped the mouse tightly, moved the cursor to the status bar, and pressed the modification key.

[Status Changed To: Resume Scheduling, Listed as Key Task of Next Stage]

8:29 PM.

The "Domestic Joint Independent Review Application" and the complete set of materials were packaged and submitted.

The raw data continued to maintain a Physics isolation state, preparing to welcome the final audit by external experts.

At the same time, in the system of the Research Support Unit, the final delivery note automatically finished refreshing.

[Core Breakthrough: Code distance 5 logical error rate is stably lower than code distance 3 in three separately recalibrated windows.]

[Core Breakthrough: Relative error suppression magnitude after scaling code distance reaches 32.2%, international public reference is approx. 3.8%.]

[Status: CLOSED]