State transitions and acquisition timing.

Question

Do triggered observations follow a changed control while untriggered reads retain the preceding result, and how do transition history, added delay and repeated acquisition affect six-state recognition? The preceding return met its trigger-status criterion but retained one wrong and ten unresolved triggered classifications. This experiment separates direct transitions from transitions preceded by a diagnostic read.

Declared method

Three separately acquired boots on the same apparatus. Each boot contains six blocks and 3,240 observations: 1,296 training observations and 1,944 held-out diagnostic observations. All values below are decimal unless explicitly prefixed. Retain control codes 90, 93, 96, 99, 102 and 105, the two-count calibration guard, operating limits, bus deadlines and restoration path from the preceding experiment.

Each block begins with 36 consecutive triggered training readings at each state, using a 1,000 microsecond delay and alternating ascending/descending state order by block. Freeze its calibration before any validation. A rejected calibration retains all raw acquisition with recognition unavailable; its outcome cannot count as successful recognition.

Each block then tests all 36 ordered pairs A → B, including the six unchanged-state pairs. Every pair has nine observations in two sequences. Both sequences begin with their own triggered A reference. The direct sequence therefore measures the first B trigger without an intervening diagnostic read.

For zero-based block b and pair position p, enumerate pair (5p + 7b) mod 36, with A its quotient by six and B its remainder. Select delay index (p + b) mod 3. If p is even, execute stages 0–3 followed by 4–8; otherwise reverse the two sequences. Every pair sees each of the six delay/order combinations once per boot, and each block contains 18 instances of each sequence order. This is a deterministic balanced schedule, not randomized sampling; the full ordering remains available for assessment.

Record actual control-selection, ADC-call and trigger-slot counters. The named delay is the wait before ADC setup, not a claim of exact selection-to-conversion latency. Stage 6 includes the preceding stage-5 diagnostic transaction and cannot be treated as a direct 1 ms transition. A 20 ms hold is a comparison condition; its physical sufficiency is not assumed.

All modes retain ADC reset/configuration, the observed status sequence and two result-pair reads. Stages 5 and 8 omit the trigger and cannot enter native recognition. The remaining 1,512 held-out triggered observations are recognized against that block's frozen bands. The 432 omitted-trigger observations retain their raw values and explicitly unavailable recognition. Hardware, operating-condition or transport failure preserves the attempted prefix and ends acquisition with the existing restoration attempt. Restore the original control after every completed block.

Predefined assessment

Require complete recovery of all 3,240 observations and all six block restorations, successful acquisition and operating checks, matching requested/initial/final control values, complete diagnostics and agreement between both result pairs. Triggered observations must retain status octets 0 → 1 → 0; omitted-trigger observations must retain 0 → 0 → 0. Missing or failed observations cannot satisfy a criterion.

Retention and update: in every one of the 216 paired trials, stage 5 must equal the preceding A reference at stage 4, and stage 8 must equal the preceding B result at stage 7. For the 180 trials with A different from B, stage 6 must differ from stage 4. The 36 unchanged-state trials impose no numerical-change requirement on a newly triggered conversion. Retention and numerical update are assessed separately from the correctness of the recognized state.

Six-state recognition: all six calibrations must be valid, and all 1,512 triggered held-out readings must be correctly recognized. Count wrong, unresolved and unavailable readings separately for every stage and selected delay. The per-boot combined criterion requires both retention/update and recognition criteria. A three-boot recognition claim requires the complete predefined sequence to qualify on all three boots; one successful boot establishes only its observed result.

Secondary comparisons examine first versus subsequent B readings, held-state references versus first-transition readings, early versus later blocks, sequence order and requested predecessor. Preserve all training extrema, raw octets, timing and operating observations. No warm-up observations are discarded, no band is widened after seeing validation, and no failed first reading is replaced with a later reading. Timing effects and drift remain hypotheses to test against these observations.

The outcomes test whether the measured results respond to the controlled trigger/change sequence. They do not independently establish analog sampling time, absolute ADC accuracy or exclusive device ownership. A separately validated observation path remains necessary for independent analog confirmation.

Results across boots

All three planned boots returned complete journals, with 3,240 observations each and 9,720 in total. Every boot met the retention/update criterion; none met the recognition or combined criterion. Each boot also includes 432 omitted-trigger controls, excluded from the triggered-recognition totals below.

The lower unresolved count in boot 2 cannot establish improved recognition: its rejected first calibration left 252 triggered readings unavailable. The three-boot series is complete. Across 4,536 scheduled triggered held-out readings, 4,260 were correct, 0 were wrong, 24 were unresolved and 252 were unavailable. The 1,296 omitted-trigger controls are separate.

Boot 1 / 43898E7FEFE4

The retention/update criterion was met; the six-state recognition criterion was not met. Candidate 43898E7FEFE4 returned all 3,240 observations. All 216 stage-5 readings retained their preceding A result and all 216 stage-8 readings retained their preceding B result. In all 180 changed-state trials, the first trigger after omission produced a value different from its A reference.

All 2,808 triggered observations, including training, recorded status octets 0 → 1 → 0. All 432 omitted-trigger observations recorded 0 → 0 → 0. Both result pairs agreed throughout, requested and returned controls matched, and all six block restorations and the final restoration reported success.

Triggered held-out recognition produced 1,497 correct, 0 wrong and 15 unresolved results out of 1,512. All six calibrations were accepted. The 432 omitted-trigger controls had recognition unavailable by design and are excluded from that denominator. Blocks 2 and 5 recognized all 252 triggered readings correctly; the remaining blocks did not meet their recognition criterion.

Boot 2 / ABC1CA660F15

Retention/update met its criterion in all 216 paired trials. All 216 stage-5 readings retained their A reference, all 216 stage-8 readings retained their B reference, and all 180 changed-state trials updated after triggering. All 2,808 triggered observations recorded status 0 → 1 → 0; all 432 omitted-trigger observations recorded 0 → 0 → 0. Both result pairs agreed throughout. Control readbacks, operating checks and all six block restorations reported success.

Triggered recognition produced 1,255 correct, 0 wrong, 5 unresolved and 252 unavailable results. The first calibration was rejected; the remaining five were accepted. Blocks 3 and 6 recognized all their triggered held-out observations correctly.

Boot 3 / 2ED4670D6C43

Retention/update met its criterion in all 216 paired trials. All 216 stage-5 readings retained the preceding A result, all 216 stage-8 readings retained the preceding B result and all 180 changed-state trials updated after triggering. The 2,808 triggered observations recorded status octets 0 → 1 → 0; the 432 omitted-trigger observations recorded 0 → 0 → 0. Both result reads agreed throughout. All control readbacks and operating checks succeeded, and all six block restorations and final restoration reported success.

Triggered recognition produced 1,508 correct, 0 wrong and 4 unresolved results. All six calibrations were accepted, so no triggered recognition was unavailable. The four unresolved observations occurred in the first two blocks; blocks 3–6 recognized all 1,008 triggered held-out observations correctly.

Boot 1 transition and timing observations

Every first direct B reading in the 180 changed-state pairs was recognized correctly. The two unresolved stage-1 readings occurred in unchanged-state pairs: 4 → 4 at ordinal 397 and 0 → 0 at ordinal 1,531. All 216 first triggered readings after omission, at stage 6, were recognized correctly.

The other 13 unresolved readings occurred in three A references, four subsequent B readings and six held-state B references. Seven unresolved readings occurred in block 1; eight occurred later, including two in block 6. All remain included in the assessment.

The selected-delay columns describe each paired trial. Only stages 1 and 5 use that selected wait; stages 0, 3, 4 and 7 always wait 20 ms, and stages 2 and 6 always wait 1 ms. Stages 5 and 8 are omitted-trigger controls and do not appear in the recognition table. These sparse counts do not establish a ranking of delays.

The programmed wait is only part of the measured interval. ADC setup and the intervening diagnostic transaction account for additional recorded time; stage 6 cannot be interpreted as a direct 1 ms transition. These counters delimit software operations and do not independently locate analog sampling.

Each unresolved value lies outside every frozen recognition band. The complete training extrema, guarded intervals, raw octets, requested states and recorded counters remain in the decoded journal. The secondary assessment additionally groups outcomes by stage, requested state, pair source and sequence order. Direct-first trials contained ten unresolved readings and omission-first trials five, from 756 triggered held-out readings in each order; this single deterministic schedule does not identify a causal order effect.

All recorded operating checks reported success. Frequency readings ranged from 1,499,971,336 to 1,500,028,995 Hz; retained thermal codes ranged from 748 to 771. The thermal codes are not independently calibrated temperatures. Acquisition took 135.169 seconds; trial start to the final recorded section took 953.468 seconds, approximately 15 minutes 53 seconds. Neither interval measures the complete operator-reported power window.

Boot 2 calibration and recognition

The first block's state-3 training interval was 2,959–3,008; state 4 was 3,005–3,040. Their guarded intervals, 2,957–3,010 and 3,003–3,042, overlap. The training intervals already overlap before the guard is added. The rejection therefore cannot be removed solely by reducing the two-count guard while retaining the observed extrema. The recorded sets of individual raw codes for those states had no identical value; the overlap concerns their intervals.

At ordinal 140, a state-3 training observation returned raw code 3,008 in both result reads. The requested control and both readbacks were 99. This was the 33rd consecutive state-3 training observation. Raw code 3,008 was also observed for a state-3 request in the preceding trigger experiment, where it produced the reported wrong identity. The recurrence is evidence about the recorded result; its physical or acquisition-related cause remains unresolved.

The rejected calibration made recognition unavailable for all 252 triggered held-out observations in block 1. All raw readings and paired diagnostics were retained. The separate 432 omitted-trigger controls account for the remainder of the 684 unavailable classifications in the full diagnostic summary.

Among changed-state pairs, 150 of the 180 first direct B readings were available for recognition and all 150 were correct; the other 30 belonged to the rejected block. Of 216 first triggers after omission, 180 were available and correct and 36 were unavailable. These unavailable outcomes remain in the assessment.

Two unresolved readings occurred in unchanged-state pairs: the first direct reading at ordinal 757 and the held-state reading at ordinal 2,077, both with A = B = 1. The other three were two A references and one subsequent B reading. No wrong identity was assigned where recognition was available.

Every cell additionally contains 12 triggered observations with unavailable recognition from block 1. Selected delay applies only to stages 1 and 5; the other stages retain their prescribed waits. Neither these counts nor the total reduction from 15 to 5 unresolved readings qualifies a superior delay or an improvement across boots.

Acquisition took 135.167 seconds; trial start to the final recorded section took 953.706 seconds, approximately 15 minutes 54 seconds. Frequency readings ranged from 1,499,969,891 to 1,500,029,660 Hz; retained thermal codes ranged from 748 to 771. These recorded intervals and uncalibrated thermal codes do not independently measure total powered duration or temperature.

Boot 3 recognition and timing

All 216 first direct B readings were correctly recognized, including the 180 changed-state pairs. Three of the four unresolved readings occurred at stage 6, the first trigger after omission; the fourth was an A reference at stage 0. Stage 6 had no unresolved readings where recognition was available in the first two boots. Its three failures in boot 3 prevent treating that sequence as a demonstrated recognition remedy.

At each unresolved stage-6 observation, the raw value differed from its preceding A reference but lay outside the frozen B band. The numerical-update requirement was therefore met while recognition remained unresolved. No first reading was replaced by a later reading.

Selected-delay columns describe the paired trials; only stages 1 and 5 use that selected wait. Stage 6 always uses a 1 ms pre-ADC wait after the preceding diagnostic transaction, while stage 0 always uses 20 ms. The three stage-6 failures cannot be interpreted as failures of a direct 1 ms transition.

Acquisition took 135.169 seconds; trial start to the final recorded section took 953.480 seconds, approximately 15 minutes 53 seconds. Frequency readings ranged from 1,499,969,891 to 1,500,028,330 Hz; retained thermal codes ranged from 746 to 770. These counters and raw thermal values do not independently measure total powered duration or temperature.

Three-boot assessment

The retention/update relation met its predefined criterion on all three boots. Across the series, all 648 A-retention checks, all 648 B-retention checks and all 540 changed-request numerical-update checks passed. Independent analog conversion freshness, absolute ADC accuracy and exclusive device ownership remain unproven.

The three-boot recognition criterion was not met. All boots contained unresolved readings, and boot 2 additionally rejected its first calibration. The series contains no wrong assigned identities where recognition was available, but this cannot qualify recognition while 24 triggered readings remained unresolved and 252 remained unavailable.

Of 540 first direct readings in changed-state pairs, 510 were available and correctly recognized; 30 were unavailable in boot 2. Considering all first direct B readings, including unchanged-state pairs, stage 1 produced 609 correct, 3 unresolved and 36 unavailable results. Stage 6, the first trigger after omission, produced the same aggregate counts, with all three unresolved outcomes in boot 3. These observations do not establish a superior sequence, causal timing effect or independent trials.

Failures in unchanged-state and held-state observations, together with the overlapping training intervals in boot 2, remain unexplained by a simple shortage of delay immediately after a requested state change. Boot 3's four later blocks passed, but later failures remained in boots 1 and 2. A general warm-up remedy is therefore unqualified. No bands were widened and no failed or unavailable observations were discarded.

Next investigation: characterize reported raw-code variation during repeated acquisition at a held request, retaining training extrema and complete acquisition diagnostics. Compare any revised acquisition or recognition rule on new held-out data. Physical variation, calibration coverage and acquisition-path effects must remain distinct hypotheses; independent analog sampling evidence is still required. The completed series does not select a revised recognition rule.

The reproducible series assessment retains per-boot outcomes, separate triggered and omitted-trigger counts, stage comparisons and the original qualification limits.

Returned evidence and provenance

Boot 1

The operator confirmed the prescribed procedure: at least five minutes powered off, one uninterrupted 30-minute boot and power off before removal. The duration is operator-reported. Two read-only acquisitions of the complete 67,108,864-byte experiment extent matched exactly. All 20,784 recorded sectors were recovered with no journal issues; 11,920 reservations remained empty.

Returned capture SHA-256: 437d75d46f411b977cc3608577f1edf0d216ce36c0140fa0453d4db3b1d1fd80

Windows reported FAT32 / Healthy / OK. Both boot files were unchanged and the examined dirty indicators were absent. Seven sectors outside the journal differed from preparation; their exact byte differences are retained in the boot-volume assessment. The writer and timing of those changes are not established. The occupied journal was preserved by read-only acquisition before preparing boot 2.

Boot 2

The prescribed procedure was at least five minutes powered off, one uninterrupted 30-minute boot and power off before removal. Actual duration and boot count were not separately reconfirmed on return or independently measured. Two read-only acquisitions of the full 67,108,864-byte extent matched exactly. All 20,784 recorded sectors were recovered with no journal issues; 11,920 reservations remained empty.

Returned capture SHA-256: c1bbb6aef88be9388d78370bee5aa2a0cb465312b0cdd4c8f6e80ba4459e0b95

Windows reported FAT32 / Warning / Full Repair Needed. The saved boot-status octet was 1 and both FAT entry-1 values were 0x7fffffff, matching the dirty indicators assessed by the existing reader. Nine sectors outside the journal differed from preparation. Both boot files, partition geometry and the complete experimental journal remained readable; boot-file contents were unchanged. The writer, timing and cause of the dirty markers are not established. No format, repair or write was performed during acquisition.

Boot 2's occupied journal was preserved before preparing boot 3.

Boot 3

The prescribed procedure was at least five minutes powered off, one uninterrupted 30-minute boot and power off before removal. Actual powered duration and boot count are not independently recorded. Two read-only acquisitions of the complete 67,108,864-byte experiment extent matched exactly. All 20,784 recorded sectors were recovered with no journal issues; 11,920 reservations remained empty.

Returned capture SHA-256: 8fa0caa2af0483c17395bbd7eeb5ea2c90dd7ddd81cf5e27f906935337eb46a9

Windows reported FAT32 / Healthy / OK. Both boot files were unchanged and the examined dirty indicators were absent. Seven sectors outside the journal differed from preparation; their byte differences remain in the boot-volume assessment. The writer and timing of those changes are not established. Acquisition was read-only. The completed third run is preserved in the return archive.

Execution path and dependencies

Existing ROM/EEPROM entry loads project-authored literal AArch64 SEIGRASM instructions. The existing launch, GPIO/I2C access, bounded PMIC acquisition, supervision, calibration, restoration and SGDATA01 SD recording remain operative. The new instructions enumerate matched transitions and select diagnostic mode and delay. Recognition excludes explicitly untriggered observations. PC tooling copies literal opcode words, places data and validates saved evidence; it does not implement target execution or invoke a foreign target compiler or assembler. Existing boot-firmware dependence remains unresolved.

All six control values, the permitted control range and hardware failure guards remain unchanged. ADC status transitions and response to control changes are operational evidence; they do not independently establish analog sampling time or absolute ADC accuracy.

The transition operation supplies sequence and sample-dispatch entries. Acquisition entry 0x3b0 extends the shared operation at two dispatch words and appends a third entry; earlier entries retain their counts and recognition behavior. The new entry retains 64-word rows and excludes diagnostic mode 2 from recognition. The PC return reader independently reconstructs scheduling, status, retention and recognition assessments from the saved journal. Configuration schema 5 retains the declared geometry and permutation. Diagnostic schema 2 is unchanged. In held-out rows, fields 5, 6 and 7 contain delay index, ordered pair identity and logical stage; training retains its existing storage-group meanings.

The maximum journal occupies 20,784 of 32,704 reserved sectors, leaving 11,920 empty reservations. Target record writing remains SEIGRASM-owned and precedes the boot partition. Earlier evidence and its frozen sources remain unchanged. Execution validation checks the full schedule, balanced coverage, mode/delay dispatch, negative-control recognition exclusion, failure restoration and connected entry-to-journal behavior under supplied devices.

SD preparation and procedure

Boot 3 of 3 was installed and readback-verified as 2ED4670D6C43. Executable instructions, configuration and the declared protocol are unchanged. Only the acquisition identity header and journal reservations differ from boot 2. All 32,704 reservations were verified empty before deployment. The existing 209-check validation record was reused after verifying all 72 source hashes. The pre-write backup exactly matched the preserved boot-2 capture. A reopened-device readback of all 67,108,864 bytes matched the prepared image. Windows reported FAT32 / Healthy / OK after installation. All three boots have returned; the declared series is complete.

Boot 3 image SHA-256: 2452b2d0e58a2b5d14d6244cd57d0855b7eb256f1d6dabaf39115993d8b5bb07

Boot 2 of 3 was installed and readback-verified as ABC1CA660F15. The executable instructions, configuration and declared protocol are unchanged from boot 1. Only the acquisition identity header and journal reservations differ. All 32,704 reservations were empty before deployment. The existing 209-check validation record was reused after verifying all 72 source hashes. The pre-write backup exactly matched the preserved boot-1 capture, and a reopened-device readback of all 67,108,864 bytes matched the new image. Windows reported FAT32 / Healthy / OK after installation. Boot 2 results are recorded above.

Boot 2 image SHA-256: 9f15f2c19e31306125d341755b23ed2a9df8d5399d3b8c21348b3cd20ff28cac

Candidate 43898E7FEFE4 was installed and readback-verified for boot 1 of 3. All 209 software checks passed against unchanged sources. A reopened-device readback of all 67,108,864 bytes matched the prepared image. The preserved pre-write backup matched the preceding returned capture exactly. Windows reported FAT32 / Healthy / OK after installation. Boot 1 has returned.

Boot 1 image SHA-256: 05fb4fd50e9308b21c66c35f23f1993a755f7d63daa38ea4cd6b9f38d8e78888

1. Keep the Pi powered off for at least five minutes before the boot. Retain the same supply and cooling.
2. Safely eject the verified SD from the PC and insert it into the powered-off Pi.
3. Power on once and leave it uninterrupted for 30 minutes. The sequence runs once, attempts restoration, records its journal and parks.
4. Power off before removing the SD and return it to the PC for read-only acquisition. The occupied journal must be preserved before preparing the next boot.

The 30-minute power window retains the preceding procedure. This candidate has fewer recorded observations and fewer journal sectors, and the first return records 953.468 seconds from trial start to the final section. The second and third boots use separately prepared identities and empty journals after each preceding return is captured.