gpt-6.2-sol high: aht-repair-complexity
Completed visible assessment, preserved as a readable check record. It is a model assessment, not a formal certificate. This document contains no private reasoning stream.
R2 passes. The local telescope, stabilization, and post-truncation merger check establish a polynomial bit bound for the modified schedule, including transmissions by multiple inverse powers. No complexity assertion about the original schedule is used.
R2
Checked statement: The modified fixed-Q schedule has polynomial bit complexity on arbitrary signed interval pairings, by the width-product and strict-closeness argument in parts (2)-(3), including the stabilization and transition facts needed from part (1).
Returned assessment: no_issue_found
Stabilization and the next selection. Normalize by putting the smaller start first; equal starts give either a translation identity or a reflection on coincident sides. The latter is trimmed by the same midpoint restriction as any overlap, omitting a fixed midpoint and discarding an empty restriction. Singletons can retain the sign obtained by composition. With Q fixed and range [c,n], an unshifted side of an eligible P starts below c: its end is at most n, so otherwise it too would be eligible. For increasing Q of step t>0, the maximal inverse exponent on a side starting at s is floor((s-c)/t)+1, and its final start is in [c-t,c-1]. For trimmed reversing Q, the entire inverse image is left of c. Thus only trimming can make a pairing eligible again. At a repeat the trimmed reflection's domain starts below c, so only its range moves. A reversing Q then produces a translation and stops the repetition. For increasing Q, a repeat with reflection constant H has range start u=floor(H/2)+1; the shift m is at least u-c+1. Normalization and restriction do not change the new constant H-m. Hence if there is another repeat, its nonnegative excess e=H-2c+2 is at most ceil(e_old/2)-1. Since the initial excess is at most 2N, the transmission count per pairing is O(1+log N), including the first transmission. Processing another pairing does not undo this stabilization.
Every other range ending at n was initially eligible by maximal width. Such a range cannot still end at n after stabilization: its start is now below c and its width never increased beyond Q's width. No unprocessed range newly reaches n. The largest other range endpoint also bounds all other domain endpoints by normalization. Thus the maximal exclusive suffix is nonempty and the retained endpoint is the maximum of c-1 and the largest other range endpoint, taking c-1 if there are no others. If Q survives, another range ends at that retained endpoint and starts strictly left of c. No terminal cleanup changes the endpoint; with no merger or deletion, the next selected pairing is different and contains, hence transmits, the shortened Q range. The equality needed below is between a post-truncation endpoint and the next pre-truncation endpoint, not between two pre-truncation endpoints.
Potential and telescope. Set X to the product of 4w over the live pairings, with empty product 1. On a constant-k block, let n_i be the endpoint at selection i and put D_i=Z-c_i+1. At the first selection w_1=D_1. At adjacent selections n_(i+1)=c_i+w_i'-1 and delta=c_i-c_(i+1)>0, giving w_(i+1)=w_i'+delta and D_(i+1)=D_i+delta. Since n never exceeds Z, w_i'<=D_i, so w_i'/w_(i+1)<=D_i/D_(i+1). Multiplying the truncation ratios, with all intervening trims only reducing X, gives X_after/X_start<=w_r'/D_r. Equality in 2w_r'<=D_r therefore suffices for halving. Deletion, including a truncation that eliminates its pairing, removes a factor of at least 4. A qualifying merger has union span width at most (w_1+t_1)+(w_2+t_2)-(t_1+t_2)=w_1+w_2. The replacement pairing width is no larger than that, hence at most 2w_1w_2, so its potential ratio is at most 1/2. The remaining moves do not increase X.
Close-count argument. For a side of width w and right endpoint b, strict closeness is w>Z-b. A restriction decreases width and does not move b right. Each inverse transmission moves a shifted side's right endpoint left, even when the inverse reverses it; normalization only swaps the side labels. Thus at constant k no individual side can become close by these moves, and no loss can be offset by a gain. With no block end, the just-truncated range is close by the telescope. If the next Q transmits it without close-count loss, its inverse image must be close. A trimmed reverser's domain has at least its own width of points to its right up to Z, so it is not close, and neither is any subinterval. Thus Q is increasing. Its range contains the close incoming side and its domain contains the close image, making both sides close before truncation and, absent a loss, after truncation.
Merger after successive transmitters. Let P be the first such transmitter after its truncation, with close range J of start j and width w. The second, Q, is increasing with period candidate t=c-a>0, and Q sends J to start j'=j-rt in its domain, so j'>=a and r>=1. P remains increasing whether one or both of its sides shift, and requires no trim or repeat. Its new normalized starts L<=R<c include j'; equality would delete an identity. The two sides of P remain close, as do Q's sides after truncation. Equal-width close sides have start difference strictly less than their width: the earlier side's right gap is at least the start difference. Both nonidentity translations are therefore periodic. The same comparison applies to the old J and the new close side at L, even though they occur at different times: both have width w, J fits within [1,Z], and L<=j'<j. Hence w>j-L. Since j-j'>=t and j>=a+t, this implies w>=t+max(0,a-L). Q's truncated endpoint still covers R+w-1 because P is another live pairing, and R>=j'>=a. The overlap of the two spans is exactly R+w-max(L,a), at least (R-L)+t. This verifies the actual merger test using t, not rt, with no assumption that only one side of P shifted.
Counting and bit cost. Within a fixed-reference block, four complete consecutive macros with neither a block end nor a close-count loss are impossible: the second and third are the successive transmitters just analyzed, and an available merger takes priority over the fourth. At most 2k close sides can be lost, so there are O(k+1) macros per block. For bookkeeping, a halving reached inside a macro can be charged at its completion, since X cannot rebound; the ending macro is counted once, and a k-drop restart or merger-only step is charged to its drop. This needs no change to the operations. Each paid reset costs a factor of at least two, so from 1<=X<=(4N)^k there are O(1+k(1+log N)) blocks, allowing a final unpaid block. All merger candidates can be scanned in pairs, and a terminal suffix can be found by a maximum-endpoint scan, without visiting individual points. Maximal powers use integer division and multiplication, not repeated shifts. There are O(k(1+log N)) transmissions per macro, with endpoint arithmetic and gcd computations on O(1+log N)-bit integers. Even if tracked, X has only O(k(1+log N)) bits. If there are no pairings, including on empty input, the remaining interval is counted directly. These bounds give polynomial bit time for the modified schedule.
Premises used
- Intervals are inclusive integer intervals; the initial N is the input endpoint bound and k is at most the input number of pairings.
- The modified schedule leaves interior static points in place, uses maximal exclusive terminal truncation, and gives qualifying periodic mergers priority at the next step.
- The elementary move semantics are those in the supplied AHT excerpt. Interpreting the runtime bound as a bound for an orbit counter uses the stated orbit-preservation properties; their general correctness is not a separate target of this audit.
Missing inputs
Scope changes
- The bound concerns the modified fixed-Q schedule only. It does not bound the exact printed cycles or verify a progress lemma for them.
- No normal-surface or certificate premise is needed for this complexity bound; the application remains conditional on those separate obligations.
Outside the audit
- The printed algorithm's progress lemmas and running time.
- The separate general correctness proofs of the elementary orbit moves, beyond the local properties used here.
- The application claim, normal-surface foundations, NP certificate premises, and NP-hardness reduction.
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.