Development progress

Snapshot as of 2026-09-09T11:51Z (UTC); updated at least daily while the evidence campaign is active. The data behind this page is committed at docs/progress/progress.json.

Dynamics builds Social Security earnings histories from the PSID, and the corrected covered-earnings series is only as credible as the reading of the source documentation beneath it. Before that series ships, every piece of documentation the construction relies on is independently verified by the three arms below; the design is pre-registered and ratifies only when they complete. (Internally: the evidence campaign behind Amendment 20 (two-arm evidence charter for the covered-earnings correction design), under review in PR #405, open; draft ratified through round 6.2; next step, the A4 evidence freeze, then the C20 ratification chain toward revision 22.)

Already built

The foundations below are complete and in the repository today; the campaign tracked on this page is the verification layer on top of them.

Verification arms

<strong>Prompt-purpose census</strong>
<span style="font-variant-numeric: tabular-nums;">
  20,815 / 20,815 &middot; 100.0%</span>
<div style="background: #2c6496; border-radius: 6px; height: 14px;
            width: 100.0%;"></div>
prompts adopted (contiguous R1–R20815 — ARM COMPLETE, terminal fold v1.97). Establishing what each of the ~21,000 remaining items in the PSID’s 1968–2023 questionnaires and codebooks is actually asking, so the model only uses variables whose meaning has been adjudicated rather than assumed. Each block runs a full pass, an independent dense audit, and a correction sweep before its entries are adopted into the governing ledger. 
<strong>A3 classification arm</strong>
<span style="font-variant-numeric: tabular-nums;">
  41,103 / 41,103 &middot; 100.0%</span>
<div style="background: #2c6496; border-radius: 6px; height: 14px;
            width: 100.0%;"></div>
queue rows final-adopted — ARM COMPLETE. Deciding how zeros and missing readings in the source data should be interpreted — a true zero, a question that was never asked, or an inapplicable route — across a 41,103-row queue of PSID variable readings, in 800-row blocks with 20% independent audits and correction sweeps. 
<strong>Q5 semantic annotation</strong>
<span style="font-variant-numeric: tabular-nums;">
  63 / 257 &middot; 24.5%</span>
<div style="background: #2c6496; border-radius: 6px; height: 14px;
            width: 24.5%;"></div>
documents sealed — 63 of the 81 annotated questionnaires (77.8%); the 257 denominator is the fixed PSID source-documentation domain: every survey instrument is annotated, and the other 176 are non-instrument files (setups, codebooks, raw data), 43 of them classed Semantic N/A by the coverage audit (not a law term). Verifying which survey questions connect to which data fields, document by document across the PSID questionnaire corpus: two independent annotators per document, then a reconciliation that seals its table of record.

Inside the numbers

The three drill-down tables below decompose each arm’s progress bar row by row. Every row was extracted from the campaign’s evidence archive and independently re-verified against the artifact named in its source column.

Purpose census, document by document

What one unit is: One rank = one row of the pinned 20,815-row residual index (e8-ops/sol-ce-purpose-residual-index-v1.txt, file SHA b35519702e6f…) = one prompt occurrence keyed psid-questionnaire-occurrence:<sha256> — a single printed questionnaire item in a sealed official PSID questionnaire/QxQ-codebook PDF (staged under ~/PolicyEngine/psid-data/documentation/capture1), located by an exact page/character-span receipt, e.g. q88:p4[2538,2708)#10bc6b307d41. What is adjudicated per rank is the prompt’s “purpose”: per sol-ce-purpose-evidence-scope-report.md, purpose “is the closed model-law classification saying which factual dimension a field_purpose_prompt establishes for O_P” — not survey designers’ motivation. Each rank is disposed either D (determined: an ordered purpose array over the closed 35-literal space, e.g. [M]) or SU (source-underdetermined, with a named condition such as U_ONTOLOGY, U_CONFLICT, or U_SCOPE). Example full row from sol-ce-purpose-ledger-pass-R6172-6576-report.md: “R6176 | Q7316 | psid-questionnaire-occurrence:091095d3… | E3.CALENDAR_EXPOSURE | D | [M] | q88:p4[2538,2708)#10bc6b307d41”. Documents are the per-wave PSID instruments: f-docs are family-interview QxQ codebooks (e.g. f94 = fam1994_QxQs.pdf) and q-docs are questionnaires (e.g. q94 = q94.pdf, “1994 Questionnaire”, per the e4-era build’s source table).

Why the denominator is 20,815: 20,815 = 21,971 − 1,156. sol-ce-purpose-residual-index-v2-report.md states the domain law verbatim: “The domain law yields 20,815, not 20,809: E ⊂ D; |C − D − E| = 21,971 − 1,156 = 20,815”, where C is the canonical corpus (sol-ce-purpose-residual-index-report.md: “The canonical corpus contains 21,971 prompts. Subtracting the same IDs yields 20,815”) and the 1,156 is the already dual-covered calibration relation from sol-ce-purpose-grand-reconcile-report.md (“Unique dual-covered prompts: 1,156” = production ranks 1–1,130 plus 26 unique rubric-sample rows; the six calibration-only exclusions are inside those 26). So the artifacts state the relation to 21,971 directly: 21,971 canonical prompts = 1,156 dual-covered in the pre-census calibration phase + 20,815 residual ranks R1–R20815 that the census adjudicates. The index was pinned 2026-08-23 (board: “PURPOSE RESIDUAL INDEX PINNED: e8-ops/sol-ce-purpose-residual-index-v1.txt (20,815 rows; R-Q SHA 858a5aa9; file SHA b35519702e6f2e2a2541c25f62878d0125f86c31b383fbc681b5369d0c272c90)”). Distinct nearby numbers, per the same artifacts: 21,099 is the established-Q fail_closed_no_rule queue (an earlier Q namespace; “Subtracting 1,156 yields 19,943 rows, not approximately 20,815” — that derivation was explicitly rejected in favor of the canonical 21,971 ordinal), and 21,153 is the evidence-scope report’s “operative burden” of underdetermined prompts. Board percentage milestones are rank_end/20,815 (e.g. R11260/20,815 = 54.10%, matching “contiguous R1–R11260 = 54.10%”).

Document Ranks Count Status Dispositions Audit Adopted
early blocks R1-R4350 (aggregate: pilot rounds r1-r6 + pre-index production blocks + 452-rank gap block, re-keyed to the pinned index) R1–R4,350 4,350 adopted 3,898 re-keyed rulings = 3,487 D / 411 SU; plus 452 gap-block rulings (D/SU split not stated) coverage audit vs the pinned index: 0 duplicates, 452 gaps in R1-R4350 -> gap block dispatched; gap block reconciled 83.6% 2026-08-23
block R4351-R4650 R4,351–R4,650 300 adopted — reconciled 270/300 = 90.000% (below cal-3 reference 93.333%); adopted as table of record under the explicit policy change 2026-08-23
block R4651-R4950 R4,651–R4,950 300 adopted — reconciled 260/300 = 86.667% (24 application failures + 16 coverage gaps) -> triggered stop-and-diagnose 2026-08-23
block R4951-R5250 (f87 opening, families B-D) R4,951–R5,250 300 adopted — reconciled 264/300 = 88.000% RECALIBRATE - the last dual-pass block, folded as coverage data 2026-08-23
block R5251-R5550 (f87 close + q87 open; first ledger-first calibration block) R5,251–R5,550 300 adopted 300 = 294 D / 6 SU (corrected block final, 31 corrections) 82/99 = 82.83% NOT CALIBRATED (ledger-first calibration) -> corrected via sweep; adopted at the v1.8 fold 2026-08-24
block R5551-R5614 (q87 completion) R5,551–R5,614 64 adopted 64/64 (61 executable + 3 UNCOVERED proposals), 0 SU dense re-calibration audit CALIBRATED 28/28 = 100% (zero contradictions) 2026-08-24
f88 R5,615–R6,171 557 adopted 557 = 484 D / 73 SU (sum of corrected block finals: R5615-5914 = 256 D / 44 SU; R5915-6171 = 228 D / 29 SU) R5915-6171 audit 93/100 = 93.00% NOT CALIBRATED (family-concentrated); R5615-5914 audit NOT CALIBRATED -> both blocks sweep-corrected; adopted at the v1.10 fold 2026-08-24
q88 R6,172–R6,632 461 adopted 461 = 440 D / 21 SU (sum: R6172-6576 corrected final 405 = 391 D / 14 SU; tail R6577-6632 = 49 D / 7 SU) R6172-6576 audit 87/92 = 94.57% NOT CALIBRATED (family-concentrated) -> corrected (75 corrections); tail audit CALIBRATED 28/28 = 100%; adopted at the v1.12 fold 2026-08-25
f89 R6,633–R7,210 578 adopted opener R6633-6903 corrected final 271 = 233 D / 38 SU; block-2 R6904-7210 pass census 307 = 259 D / 48 SU, adopted with the single R7041 -> U_CONFLICT correction opener audit 118/150 = 78.67% NOT CALIBRATED -> corrected (118 corrections); block-2 CALIBRATED 99.32% (145/146) 2026-08-25
q89 R7,211–R7,558 348 adopted 348 = 313 D / 35 SU (corrected block final; 60 routing-only + 10 semantic+routing corrections) 113/123 = 91.87% NOT CALIBRATED -> corrected; adopted at the v1.18 fold 2026-08-25
f90 R7,559–R8,084 526 adopted opener R7559-7820 corrected 262 = 229 D / 33 SU; block-2 R7821-8084 = 225 D / 39 SU (sum 454 D / 72 SU) opener audit 116/121 = 95.87% NOT CALIBRATED (family-concentrated) -> corrected; block-2 CALIBRATED 99.08% 2026-08-26
q90 R8,085–R8,552 468 adopted B+C R8085-8295 corrected census 211 = 173 D / 38 SU; D-M R8296-8552 corrected 257 = 182 D / 75 SU (sum 355 D / 113 SU) B+C dual audits split -> geometry adjudication CALIBRATED 85/86 = 98.84%; D-M corrected under the geometry law 2026-08-26
f91 R8,553–R8,974 422 adopted 422 = 368 D / 54 SU (corrected final; 28 sealed semantic corrections) 166/194 = 85.57% NOT CALIBRATED -> corrected; adopted at the v1.26 fold 2026-08-28
q91 R8,975–R9,571 597 adopted opener R8975-9245 corrected 271 = 261 D / 10 SU; block-3 R9246-9571 corrected 326 = 311 D / 15 SU (sum 572 D / 25 SU) opener audit 85/90 = 94.44% NOT CALIBRATED (concentrated defects); block-3 audit 96.55% NOT CALIBRATED (transition-date family) -> both corrected 2026-08-28
f92 R9,572–R9,795 224 adopted 224 = 193 D / 31 SU (corrected final; sweep found 12 contradictions, 4 beyond the audit) 100/108 = 92.59% NOT CALIBRATED -> corrected; adopted at the v1.32 fold 2026-08-28
q92 R9,796–R10,217 422 adopted 422 = 405 D / 17 SU (corrected final; UNCOVERED=0) 140/145 = 96.55% numeric pass but all six SEE-router checkpoint rows misruled -> NOT CALIBRATED -> corrected; adopted at the v1.33 fold 2026-08-28
q93 R10,218–R10,749 532 adopted opener R10218-10457 corrected 240 = 222 D / 18 SU; block-7 R10458-10749 corrected 292 = 269 D / 23 SU (sum 491 D / 41 SU) opener audit 79/89 = 88.76% NOT CALIBRATED; block-7 audit 95/99 = 95.96% NOT CALIBRATED (overlap/reclassification gateway family) -> both corrected 2026-08-29
f94 R10,750–R11,260 511 adopted block-1 R10750-11039 corrected 290 = 271 D / 19 SU; block-2 R11040-11260 corrected 221 = 198 D / 23 SU (sum 469 D / 42 SU) block-1 audit 132/139 = 94.96% NOT CALIBRATED (family-concentrated); block-2 audit 96/112 = 85.71% sample / 203/221 = 91.86% guard sweep NOT CALIBRATED -> both corrected 2026-08-30
q94 (B+C opener block) R11,261–R11,576 316 adopted corrected final 316 = 303 D / 13 SU (UNCOVERED=0; supersedes the provisional pass census 316 = 301 D / 15 SU with 221 ledger-matched / 95 UNCOVERED) 241/288 = 83.68% sample / 268/316 = 84.81% full NOT CALIBRATED - central simultaneity frame REVERSED (q94 has no simultaneity sentence) -> frame-reversal correction sweep applied; corrected final adopted at the v1.44 fold, q94 B+C CLOSED, contiguous R1-R11576 = 55.61% 2026-08-30
q94 D+E (block 4) R11,577–R11,894 318 adopted 304 D / 14 SU dense audit CALIBRATED 272/272 sample and 318/318 full = 100.00% — the campaign’s first calibrated block; folded v1.45 with no sweep 2026-08-30
q94 tail + f95 complete + q95 B opener (block 5) R11,895–R12,163 269 adopted 236 D / 33 SU (pass census) 96.90% sample / 97.40% full NOT CALIBRATED (all 7 contradictions in f95) -> f95-focused sweep 269/269; folded v1.50 2026-08-30
q95 C+D (block 6) R12,164–R12,398 235 adopted 221 D / 14 SU 91.26% sampled NOT CALIBRATED (position family + [C,H] composites) -> sweep with 11 corrections; folded v1.53 2026-08-30
q95 tail + f96 complete (block 7) R12,399–R12,690 292 adopted 253 D / 39 SU (corrected) 73.72% sample / 80.82% full NOT CALIBRATED — worst E4 block (41/56 defects in f96 B/S) -> sweep-r2 292/292 with 56 corrections; folded v1.55 2026-08-30
q96 B+C opener (block 8) R12,691–R12,922 232 adopted 219 D / 13 SU (corrected) 100.00% semantic (103/103 sample, 232/232 full) but 90.29% sample / 95.69% full route-completeness NOT CALIBRATED -> route-cure sweep (ten route repairs, no result changes); folded v1.57 2026-08-30
q96 D–L (block 9; E4 tranche close) R12,923–R13,210 288 adopted 268 D / 20 SU (corrected) NOT CALIBRATED: 6 semantic contradictions in 2 families + 4 route recoveries wrong; era-generic 29/29 semantic PASS but receipt-completeness FAIL; pass vector never serialized -> full re-serialization sweep 288/288; folded v1.59; E4 tranche closed 2026-08-31
f97 + q97 B opener (block 1) R13,211–R13,490 280 adopted 260 D / 20 SU (corrected) 62.90% NOT CALIBRATED — the campaign’s worst (f97-concentrated) -> sweep with 71 corrections; folded v1.62 2026-08-31
q97 C+D (block 2) R13,491–R13,723 233 adopted 220 D / 13 SU CALIBRATED 72/72 sample = 100.00%; 232/233 full; folded v1.63 directly from the pass 2026-08-31
q97 tail + f99 first contact (block 3) R13,724–R14,033 310 adopted 257 D / 53 SU 116/127 = 91.34% NOT CALIBRATED -> sweep applying the 23-row manifest, 310/310; folded v1.65 2026-09-01
q1999 B+C opener (block 4) R14,034–R14,266 233 adopted 220 D / 13 SU CALIBRATED 68/68 = 100.00% on all three measures — a first-contact document; folded v1.66 2026-09-01
q1999 D+E (block 5) R14,267–R14,499 233 adopted 220 D / 13 SU CALIBRATED 69/69 = 100.00%; folded v1.67 2026-09-01
q1999 G–R (block 6) R14,500–R14,779 280 adopted 230 D / 50 SU 97.14% sample / 98.93% full NOT CALIBRATED on family concentration only; closed 3-row manifest adopted directly; folded v1.68 2026-09-01
f2001 + q2001 B (block 7) R14,780–R15,059 280 adopted 226 D / 54 SU (corrected) 86.40% semantic / 83.20% route NOT CALIBRATED (f2001-concentrated) -> sweep: 159-row f2001 replay, 24 corrections; folded v1.70 2026-09-01
q2001 C+D (block 8) R15,060–R15,291 232 adopted 219 D / 13 SU CALIBRATED 72/72 = 100.00%; 0 contradictions in 232 rows; folded v1.71 2026-09-01
q2001 E/G/P/K/L/R (block 9; late-E4 tranche close) R15,292–R15,624 333 adopted 280 D / 53 SU (corrected) 92.86% sampled NOT CALIBRATED, but the full-domain replay closed a 17-row manifest -> adopted with corrections under the closed-manifest precedent; folded v1.72 2026-09-01
fam2003 + q2003 BC/DE/G (E5 block 1) R15,625–R15,883 259 adopted 224 D / 35 SU (corrected) 90.00% sample / 92.66% full NOT CALIBRATED (q2003 G concentrated) -> sweep CURED, 259-row replay with 22 E5-rule applications; folded v1.75 2026-09-01
q2003 P/KL/R + fam2005 + q2005 BC/DE (E5 block 2) R15,884–R16,138 255 adopted 194 D / 61 SU (corrected) 93.73% full strict but the sample fails the route/strict limbs, NOT CALIBRATED -> sweep 255/255, 16 substantive corrections (2 q2003 + 14 outside q2003); folded v1.77 2026-09-02
q2005 G/P/KL/R + fam2007 BCDE/F/G (E5 block 3) R16,139–R16,400 262 adopted 203 D / 59 SU CALIBRATED 146/146 = 100.00% on both limbs; folded v1.78 2026-09-02
fam2007 P/KL/R/IO + q2007 BC/DE/G (E5 block 4) R16,401–R16,729 329 adopted 247 D / 82 SU 93.98% strict sample NOT CALIBRATED (family-concentrated); closed 10-row manifest adopted; folded v1.79 2026-09-02
q2007 P/KL/R + fam2009 + q2009 BC/DE (E5 block 5) R16,730–R17,022 293 adopted 240 D / 53 SU NOT CALIBRATED on the receipt-strict limb only (165/165 semantic, 292/293 route); one route correction; folded v1.80 2026-09-02
q2009 G/R/P/KL (E5 block 6) R17,023–R17,308 286 adopted 233 D / 53 SU 97.32% sample / 98.60% full NOT CALIBRATED on family concentration; 4-row manifest adopted; folded v1.81 2026-09-02
fam2011 + q2011 BC/DE (E5 block 7) R17,309–R17,585 277 adopted 230 D / 47 SU (corrected) 77.97% strict sample NOT CALIBRATED (q2011 whole-roster EHC routing family) -> regeneration sweep-r2, 277/277; folded v1.83 2026-09-03
q2011 F/G/R/P/KL (E5 block 8; the tranche’s last block) R17,586–R17,888 303 adopted APPLY 226 / UNCOVERED 77 (corrected census) 92.04% sample NOT CALIBRATED; closed 16-row manifest across six families adopted per the closed-manifest precedent; folded v1.84 — E5 tranche complete, contiguous R1–R17888 2026-09-03
fam2013 complete + q2013 BCDE/G (final-tranche block 1) R17,889–R18,236 348 adopted 348/348 regenerated (sweep) pass 348/348 -> dense audit NOT CALIBRATED both limbs (59-row manifest, seals regenerate) -> regeneration sweep 348/348 byte-exact envelope; folded v1.87 2026-09-04
q2013 R/P/KL + fam2015 complete (final-tranche block 2) R18,237–R18,526 290 adopted APPLY 217 / UNCOVERED 73 pass recovered byte-exact from durable scratch -> dense audit CALIBRATED 99.37% sample / 99.66% full (one-row manifest R18462); folded v1.88 2026-09-04
q2015 BCDE/G/R/P (final-tranche block 3) R18,527–R18,838 312 adopted — pass 312/312 -> dense audit CALIBRATED 100% sample and full vector (empty contradiction set; 24/24 receipts byte-exact); folded v1.89 2026-09-04
q2015 KL + q2017 complete (final-tranche block 4) R18,839–R19,111 273 adopted — pass 273/273 -> dense audit CALIBRATED (sampled 144/144; full-vector replay 273/273; row contradictions 0; family concentration none); folded v1.90 2026-09-04
q2019 complete + q2021 A (final-tranche block 5) R19,112–R19,384 273 adopted — pass staged -> dense audit CALIBRATED 100% (sample 144/144; 273/273 rows examined); folded v1.91 (terminal identity 317,298 B, SHA 014aeb13…) 2026-09-04
q2021 BC (final-tranche block 6) R19,385–R19,749 365 adopted APPLY 267 / UNCOVERED 98 (12 proposal entries) pass 365/365 -> dense audit CALIBRATED with a closed three-row manifest (R19623–R19625, BC60A over-fire -> ALL.DIRECT_EXPOSURE; eight of nine charges refuted; EHC recall 25/25); folded v1.92 2026-09-05
q2021 G (final-tranche block 7) R19,750–R20,006 257 adopted APPLY 203 / UNCOVERED 54 (8 proposal entries) pass 257/257 -> dense audit CALIBRATED 163/163, zero-row manifest (fifth consecutive calibrated audit); EHC fired 18 times and was upheld 18/18 on precision and recall; folded v1.93 2026-09-05
q2021 P/KL/IMMIG (final-tranche block 8) R20,007–R20,311 305 adopted 305 ranks in two passes: R20007–R20247 APPLY 153 / UNCOVERED 88; R20248–R20311 APPLY 27 / UNCOVERED 37 q2021 §P audit (136/241 sampled = 56.4%; manifest rebuilt in both directions with zero discrepancies; era discipline clean) folded v1.94 under a ‘block closed’ heading that covered only the §P part — an orchestrator scope error (brief cut from the per-section table); the KL+IMMIG remainder (64 ranks) ran as an 8b completion pass, audited CALIBRATED 64/64 = 100.00% with zero corrections, and folded v1.95 2026-09-05
q2021 ADDRPAYMENT + q2023 BC (final-tranche block 9) R20,312–R20,618 307 adopted APPLY 270 / UNCOVERED 37 (closes q2021 incl. ADDRPAYMENT; opens q2023 §BC) pass 307/307 -> dense audit CALIBRATED: 161/307 sampled (52.4%), agreement 161/161 = 100.00%, correction manifest empty of rows (three documentary items); adopted as issued; folded v1.96 (99.05%). One 63-character rubric digest in the pass’s assertion object traced to the dispatch brief -> standing correction adopted (recompute every emitted digest from the named file) 2026-09-05
q2023 G/P/KL (final-tranche block 10) R20,619–R20,815 197 adopted APPLY 169 / UNCOVERED 28 (G 156, P 17, KL 24; by_construction_exclusions = [Q21822]); the corpus closes at R20815 pass 197/197 (APPLY 169 / UNCOVERED 28) -> dense audit CALIBRATED, reported as 102/102 = 100% (51.8% sampled) with 101 distinct ranks (one rank counted twice; 51.3% of ranks); DISAGREE set empty; correction manifest empty of rows and items; charge 9 confirmed the Poppler pdftotext 26.04.0 pin and argument equivalence over all 523 pages of q2023; charge 10 upheld the pass’s form-feed frame-recipe correction with no row moved; charge 8 swept all 20,815 index lines end to end with zero gaps or duplicates; the ninth consecutive CALIBRATED block; folded v1.97 (379,534 B / ebe3d2b9…) — the corpus closes at R20815 2026-09-05

A3 classification, block by block

What one unit is: One queue row is one compiled-variable “family” from the A3 document-local review queue: a single PSID codebook value-meaning (e.g. rank 1’s “Major assignment; probable error of greater than $300 or 10 percent of assignment value (whichever is greater)”, per sol-ce-a3-reconcile-1-300-report.md’s scope receipts, which identify each rank by stage/global/lexeme/meaning) grouping all uncovered occurrences of that meaning within one document. Per sol-ce-a3-queue-characterization.md the queue is “41,103 families / 86,840 uncovered occurrences (mean 2.11, max 38; 31,588 single-occurrence rows). All rows source_document_count = 1 (document-LOCAL…)”. Each rank is adjudicated to a missingness disposition — per sol-ce-a3-reconcile-1-300-report.md: “T = true/missing; U = underdetermined; F = false/nonmissing” — with a fourth outcome X/UNCLASSIFIED (fail-closed quarantine where O>0 or R=0) adopted from block 24001 onward under the v1.14 X-vocabulary ruling (“block finals are now four-set T/U/F/UNCLASSIFIED partitions”).

Block Final census Audit Status
1–800 T 729 / U 24 / F 47 — final
801–1,600 T 28 / U 759 / F 13 — final
1,601–2,400 T 81 / U 686 / F 33 — final
2,401–3,200 T 545 / U 69 / F 186 — final
3,201–4,000 T 568 / U 24 / F 208 — final
4,001–4,800 T 563 / U 26 / F 211 — final
4,801–5,600 T 545 / U 61 / F 194 — final
5,601–6,400 T 665 / U 92 / F 43 — final
6,401–7,200 — — unresolved
7,201–8,000 T 673 / U 86 / F 41 — final
8,001–8,800 T 578 / U 62 / F 160 — final
8,801–9,600 T 102 / U 22 / F 676 — final
9,601–10,400 T 701 / U 96 / F 3 — final
10,401–11,200 T 408 / U 391 / F 1 — final
11,201–12,000 T 503 / U 98 / F 199 CLEAN 154/154 (contradictions 0) final
12,001–12,800 T 482 / U 72 / F 246 CLEAN 152/152 (library-defect set ∅) final
12,801–13,600 T 784 / U 10 / F 6 CLEAN 160/160 (contradictions ∅) final
13,601–14,400 T 665 / U 131 / F 4 CLEAN 147/147 covered predictions (100%) final
14,401–15,200 T 323 / U 54 / F 423 CLEAN 150/150 (100%) final
15,201–16,000 T 625 / U 126 / F 49 CLEAN 148/148 (100%) (v2 audit) final
16,001–16,800 T 687 / U 38 / F 75 CLEAN 154/154 (100%) final
16,801–17,600 T 800 / U 0 / F 0 CLEAN 160/160 (100%) final
17,601–18,400 T 800 / U 0 / F 0 CLEAN 160/160 (100%) final
18,401–19,200 T 535 / U 222 / F 43 LIBRARY DEFECT — 156/158 = 98.73% (2 contradictions) final
19,201–20,000 T 743 / U 41 / F 16 LIBRARY DEFECT — 160/161 = 99.38% (1 contradiction) final
20,001–20,800 T 699 / U 100 / F 1 CLEAN 148/148 (no library-class defect) final
20,801–21,600 T 706 / U 91 / F 3 NOT CLEAN — 3 library-class defects; 141/144 agree final
21,601–22,400 T 666 / U 134 / F 0 NOT CLEAN — 7 library-class defects; 141/148 agree (frontier paused: 4.7% > bound) final
22,401–23,200 T 549 / U 248 / F 3 CALIBRATED — 40% audit, 264/264 agree, 0.00% contradictions final
23,201–24,000 T 686 / U 114 / F 0 v1 NOT CALIBRATED 65.69%; v2 NOT CLEAN 154/160 agree final
24,001–24,800 T 738 / U 59 / F 0 / X 3 NOT CLEAN — 154 agreements / 5 claimed defects (sweep: 4 audit errors, 1 real) final
24,801–25,600 T 217 / U 530 / F 2 / X 51 NOT CLEAN 151/160 = 94.38% (corrected 150/160 after 2 claims overturned) final
25,601–26,400 T 416 / U 126 / F 127 / X 131 NOT CLEAN 110/163 = 67.48% (v2; corrected score 148/163 = 90.80%, OFUM claim SUSTAINED) final
26,401–27,200 T 752 / U 47 / F 1 / X 0 CLEAN 161/161 = 100.00% (first perfect audit of the arm) final
27,201–28,000 T 338 / U 15 / F 0 / X 447 NOT CLEAN 160/165 = 96.97% (27503 claim overturned by sweep) final
28,001–28,800 T 205 / U 294 / F 38 / X 263 NOT CLEAN 110/162 = 67.90% final
28,801–29,600 T 314 / U 265 / F 45 / X 176 NOT CLEAN 145/162 = 89.51% final
29,601–30,400 T 0 / U 691 / F 0 / X 109 NOT CLEAN 139/160 = 86.88% final
30,401–31,200 T 0 / U 675 / F 0 / X 125 NOT CLEAN 136/160 = 85.00% final
31,201–32,000 T 437 / U 160 / F 3 / X 200 NOT CLEAN 136/166 = 81.93% final
32,001–32,800 T 290 / U 60 / F 0 / X 450 CONTRADICT 77/161 = 47.83% (audit’s 69-row over-quarantine claim later OVERTURNED by sweep) final
32,801–33,600 T 215 / U 47 / F 0 / X 538 CONTRADICT 147/166 = 88.55% class/decision (sweep: SUSTAIN, audit vindicated 166/166) final
33,601–34,400 T 484 / U 70 / F 1 / X 245 CONTRADICT 130/162 = 80.25% agreement (32/162 contradictions; sweep SUSTAIN 32/32) final
34,401–35,200 T 59 / U 135 / F 183 / X 423 NOT CLEAN 145/180 = 80.56% final
35,201–36,000 T 82 / U 14 / F 677 / X 27 NOT CLEAN 181/182 = 99.45% -> patch-sweep OVERTURNED the audit at 35251: CLEAN 182/182 (first fully clean A3 block); library v1.22 final
36,001–36,800 T 15 / U 40 / F 636 / X 109 CLEAN 292/292 = 100.00% at audit (no patch-sweep needed); library v1.23 final
36,801–37,600 T 12 / U 14 / F 615 / X 159 NOT CLEAN 204/225 = 90.67% (21 contradictions, minted-class F on incomplete construction frames) -> patch-sweep SUSTAINED 18/21 + overturned 3, corrected 117 over-fires; library v1.24 final
37,601–38,400 T 2 / U 8 / F 376 / X 414 NOT CLEAN 99.46% (single dispute 37821) -> patch SUSTAINED the audit + one homolog 38337; library v1.25 final
38,401–39,200 T 5 / U 34 / F 625 / X 136 NOT CLEAN 90.62% / 92.88% (57 contradictions in five families, zero predicate defects) -> patch-sweep re-ruled all 57; library v1.26 final
39,201–40,000 T 15 / U 9 / F 465 / X 311 NOT CLEAN 78.97% decision / 65.64% full (253-row over-quarantine) -> patch SUSTAIN 67/67 + 3 amendments, 154 decision corrections; library v1.27 (frame_buildability_guard) final
40,001–41,103 T 457 / U 59 / F 373 / X 214 NOT CLEAN 282/297 = 94.95% (15 contradictions + the 40819 dispute) -> patch SUSTAIN 15/15 + 48 sweep additions; TERMINUS CONFIRMED; library v1.28 — queue complete final

Q5 annotation, sealed documents

What one unit is: Sealing one document is the Q5 arm’s per-document closure. Two independent lanes (A and B) each annotate every extracted questionnaire occurrence in one PSID document with semantic-binding arrays, source-only; per the cal2 law ruling on the board, pre-freeze Q5 scope is ‘binding arrays only (NE universally), one-document dual blocks’. A reconcile lane then aligns the two lane tables row-by-row, resolves every disagreement against the PSID source PDF (dispositions tallied AB/A/B/SRC), runs an agreed-row seam audit, and emits a canonical table of record — compact sorted-key JSON, one row per occurrence, with a pinned SHA-256 (e.g. sol-ce-q5-doc-1991f-reconcile-report.md: ‘Result: reconciliation COMPLETE; canonical annotation SEALED’). The ladder is defined in sol-ce-q5-cal2-reconcile-report.md: ‘Dual independent annotation; reconcile all disagreements; 10% agreed-row seam audit’, with promotion requiring ‘Two consecutive clean blocks’. SEALED is therefore distinct from the clean-block verdict: every one of the 33 sealed documents so far also ruled RECALIBRATE (most FAIL/RECALIBRATE — raw lane agreement below the clean bar), so the ladder has stayed at one document per dual-annotation block with clean streak 0 throughout.

Document Rows sealed Verdict Sealed
1969 family questionnaire (q69) 268 SEALED / RECALIBRATE 2026-08-21
1970 family questionnaire (q70) 263 SEALED / RECALIBRATE 2026-08-22
1971 family questionnaire (q71) 252 SEALED / RECALIBRATE 2026-08-23
1972 family questionnaire (q72) 181 RECALIBRATE 2026-08-23
1973 family questionnaire (q73) 277 RECALIBRATE 2026-08-23
1974 family questionnaire (q74) 243 COMPLETE / RECALIBRATE 2026-08-23
1975 family questionnaire (q75) 346 COMPLETE / RECALIBRATE 2026-08-23
1976 family questionnaire (q76) 453 COMPLETE / RECALIBRATE 2026-08-23
1977 family questionnaire (q77) 357 RECALIBRATE 2026-08-23
1978 family questionnaire (q78) 362 COMPLETE / RECALIBRATE 2026-08-23
1979 family questionnaire (q79.pdf) 281 COMPLETE / RECALIBRATE 2026-08-23
1980 family questionnaire (q80) 444 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-23
1981 family questionnaire (q81) 468 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-23
1982 family questionnaire (q82) 466 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-23
1983 family questionnaire QxQ companion (fam1983_QxQs.pdf) 1,025 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-24
1983 family questionnaire (q83.pdf) 501 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-24
1984 family questionnaire (q84) 88 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-24
1984 family questionnaire QxQ companion (fam1984_QxQs.pdf) 3,125 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-24
1985 family questionnaire (q85) 306 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-25
1985 family questionnaire QxQ companion (fam1985_QxQs) 5,634 COMPLETE / Q5 FAIL / RECALIBRATE 2026-08-25
1986 family questionnaire QxQ companion (fam1986_QxQs) 1,815 COMPLETE / SEALED / FAIL — RECALIBRATE 2026-08-25
1986 family questionnaire (q86) 390 COMPLETE / RECALIBRATE 2026-08-25
1987 family questionnaire QxQ companion (fam1987_QxQs) 1,741 COMPLETE / SEALED / Q5 FAIL / RECALIBRATE 2026-08-25
1987 family questionnaire (q87) 435 COMPLETE / RECALIBRATE 2026-08-25
1988 family questionnaire QxQ companion (fam1988_QxQs) 1,830 COMPLETE / SEALED / Q5 FAIL / RECALIBRATE 2026-08-25
1988 family questionnaire (q88) 1,847 COMPLETE / SEALED / Q5 FAIL / RECALIBRATE 2026-08-25
1989 family questionnaire QxQ companion (fam1989_QxQs) 1,875 COMPLETE / SEALED / Q5 FAIL / RECALIBRATE 2026-08-25
1989 family questionnaire (q89) 951 FAIL / RECALIBRATE (table complete and reproduced) 2026-08-26
1990 family questionnaire QxQ companion (fam1990_QxQs) 1,723 COMPLETE / SEALED / FAIL / RECALIBRATE 2026-08-26
1990 family questionnaire (q90.pdf) 1,256 COMPLETE (seal) / FAIL / RECALIBRATE 2026-08-27
1991 family questionnaire QxQ companion (fam1991_QxQs.pdf) 1,476 COMPLETE / SEALED / FAIL / RECALIBRATE 2026-08-27
1991 family questionnaire (q91.pdf) 2,474 FAIL / RECALIBRATE (annotation complete and sealed) 2026-08-29
1992 family questionnaire QxQ companion (fam1992_QxQs.pdf) 1,158 RECALIBRATE (annotation complete and sealed) 2026-08-30
1992 family questionnaire (q92.pdf) 2,302 RECALIBRATE (table of record issued; binding-law ruling in force) 2026-08-30
1993 family questionnaire (q93.pdf) 7,315 RECALIBRATE (table of record structurally all-pass) 2026-08-31
1994 family questionnaire QxQ companion (fam1994_QxQs.pdf) 2,500 RECALIBRATE (both transports authenticated without repair) 2026-08-31
1994 family questionnaire (q94.pdf) 2,813 RECALIBRATE (2,509 bound / 304 empty; zero unresolved) 2026-09-01
1995 family questionnaire QxQ companion (fam1995_QxQs.pdf) 884 RECALIBRATE 2026-09-01
1995 family questionnaire (q95.pdf) 7,145 RECALIBRATE (largest single-document seal at the time) 2026-09-01
1996 family questionnaire QxQ companion (fam1996_QxQs.pdf) 363 RECALIBRATE (clean-block FAIL; complete reconciled table) 2026-09-01
1996 family questionnaire (q96.pdf) 7,147 RECALIBRATE (raw agreement 44.90%; all divergent families source-resolved) 2026-09-02
1997 family questionnaire QxQ companion (fam1997_QxQs.pdf) 450 RECALIBRATE (dual-serialized; seal ratified against the campaign-chain reconcile) 2026-09-02
1997 family questionnaire (q97.pdf) 7,628 RECALIBRATE (dual-serialized; lane B recovered by dual-serialization cross-recovery) 2026-09-02
1999 family questionnaire QxQ companion (fam1999_QxQs.pdf) 282 RECALIBRATE (103/282 arrays diverged; all source-resolved) 2026-09-02
1999 family questionnaire (q1999.pdf) 8,498 RECALIBRATE (largest table of record: 8,498 rows) 2026-09-02
2001 family questionnaire QxQ companion (fam2001_QxQs.pdf) 674 RECALIBRATE (clean streak 0/2) 2026-09-03
2001 family questionnaire (q2001.pdf) 2,825 RECALIBRATE (neither lane canonical; resolved under the 1992q standard) 2026-09-03
2003 family questionnaire QxQ companion (fam2003_QxQs.pdf) 370 RECALIBRATE (230/370 occurrences diverged; all source-resolved) 2026-09-03
2003 family questionnaire (q2003.pdf) 1,225 RECALIBRATE (agreement 480/1,225; all divergences source-resolved) 2026-09-04
2005 family questionnaire QxQ companion (fam2005_QxQs.pdf) 269 lane A prevailing (174 of 194 divergences to A; common-mode 1/269) — first Opus/Opus pair 2026-09-04
2005 family questionnaire (q2005.pdf) 1,212 RECALIBRATE (neither lane canonical; common-mode 0/516) 2026-09-04
2007 family questionnaire QxQ companion (fam2007_QxQs.pdf) 952 RECONCILED — lane B prevails (917/952 as submitted; common-mode 7/389) 2026-09-04
2007 family questionnaire (q2007.pdf) 1,605 lane B adopted unmodified as the table of record (1,605/1,605) 2026-09-04
2009 family questionnaire QxQ companion (fam2009_QxQs.pdf) 171 both lanes rejected; the reconciled table stands (common-mode 5/171 = 2.9%, the highest yet) 2026-09-04
2009 family questionnaire (q2009.pdf) 2,282 lane A’s coordinate frame upheld as the base; lane B refuted on its central doctrine (2,509 minted node ids) 2026-09-04
2013 family questionnaire QxQ companion (fam2013_QxQs.pdf) 1,320 lane B substantially upheld (551 of 610 contested rows) and adopted as the base with 82 corrections; lane A refuted; both refuted on the aggregate class 2026-09-04
2019 family questionnaire (q2019.pdf) 1,882 each lane reversed on exactly one axis and upheld on the other (133 + 113 = the 246 contested rows); zero minted identifiers; common-mode 0/1,636 2026-09-04
1968 family questionnaire QxQ companion (fam1968_QxQs.pdf) 156 adjudication of the contested set: 38 rows with both lanes upheld, 69 lane A upheld, 19 lane B upheld; 156 rows = seal; zero minted identifiers 2026-09-04
1969 family questionnaire QxQ companion (fam1969_QxQs.pdf) 159 lane A with exactly 7 row replacements after an 8-block adjudication (confirmed common-mode 0/85 by the per-row method; three exposures recorded on the seal) 2026-09-05
1976 family questionnaire QxQ companion (fam1976_QxQs.pdf) 385 all 224 agreed rows audited in three strata and upheld (common-mode 0/224, or 5/224 under the stricter reading); reading-invariant, no Q5-status pass asserted 2026-09-05
1979 family questionnaire QxQ companion (fam1979_QxQs.pdf) 473 revision 2 after two of its four verdicts were adversarially overturned; reading-invariant with three exposures recorded; survived the adversarial seal review 2026-09-05
1981 family questionnaire QxQ companion (fam1981_QxQs.pdf) 449 revision 2 (six rows and one count corrected under a declared adversarial pass); both lanes and the table write zero component-side sentinels; survived the adversarial seal review 2026-09-05
1977 family questionnaire QxQ companion (fam1977_QxQs.pdf) 432 revision 2; states the containment ground independently (the strongest of the four seals under adversarial review) 2026-09-05

Development scorecard

Eight pre-registered gates; all seven locked gates have a registered first pass; 48 one-shot candidate runs on the public record.

Every candidate model is registered before its single scored run, and every run — pass or fail — is committed. The failures are part of the record.

Gate What it tests Candidates First pass
Earnings-history credibility The backcast earnings process scores against held-out PSID panel moments and geometry blocks, judged against the same-panel noise floor. 13 Candidate 13 (rank_knn_v5, 2026-07-07): 5/5 seeds geometry, 4/5 battery, C2ST 0.523 ≤ 0.53 — after 12 committed failing runs.
Family transitions (marital and fertility) Marriage, divorce, remarriage, widowhood, occupancy, cohort nuptiality, and fertility transitions scored against the PSID marital and childbirth history files. 16 Candidate 16 (hazard_v16, 2026-07-09): 4/5 seeds across all 46 gated cells — after 15 committed failing runs.
Household composition Who lives with whom: coresidence stocks and wave-to-wave transitions from the PSID relationship matrix — spouse/partner, parent, child, grandchild, household size, multigenerational entry and exit. 9 Candidate 9 (hazard_v9, 2026-07-11): 4/5 seeds — after 8 committed failing runs.
Marriage × earnings joint Who marries whom by earnings — assortative mating, earnings-conditional marriage hazards, and couple shared-earnings shape, on which spousal and survivor benefit levels depend. 2 Candidate 2 (hazard_v2, 2026-07-12): 4/5 seeds.
Disability Work-limitation incidence and recovery hazards, disabled-occupancy prevalence, and near-retirement disability exit, anchored against SSA Disability Annual Statistical Report tables. 1 Candidate 1 (2026-07-12): 5/5 seeds, all 12 gated cells — passed on the first attempt.
Representative-frame transport The PSID-estimated generators transported onto the certified CPS representative frame: 44 gated cross-sectional cells plus SSA administrative margins. 4 Candidate 4 (2026-07-13): 43/43 cells in-band on 5/5 seeds — after 3 failing candidates.
Temporal-holdout projection drift The projection engine fits on data through 2014 only, then projects and is scored on the held-out 2015–2022 window — a genuine out-of-time test. 3 Candidate 3 (2026-07-24): 4/5 seeds across the 11 gated cells.
Forward projection Near-term components resolve against future administrative publications on named resolution rules — scored when the numbers come out. — — (thresholds not yet locked)

Comparisons against the incumbent models

Direct number-for-number comparisons against the incumbent models are registered as a benchmark inventory that fills in as modules land — shown here before launch, deliberately.

  • MINT / CBO / SSA benchmark inventory (tranche 2) — 2 of 60 rows computable. 60 registered comparison rows against published MINT (19), CBO (30), and SSA Statistical Supplement (11) figures; 58 rows await the modules that compute them. Each row is pinned to the captured source file by hash. (source)
  • Cross-sectional incumbent replacement (Enhanced CPS) — candidate beats incumbent. Archived scorecard: a Microcosm-US candidate against the certified Enhanced CPS 2024 production incumbent on 3,704 calibration targets — candidate loss 0.228 vs incumbent 1.405 (holdout 0.038 vs 0.317). Publication pending in microcosm-benchmarks. (source)
  • DYNASIM output comparisons — not yet. No quantitative DYNASIM-output rows exist yet; today’s DYNASIM comparison is the capability table below. Quantitative rows begin at the projection and benefit stages.

Capability scorecard vs DYNASIM

Component-by-component against DYNASIM, the main non-governmental dynamic benchmark (from the public record; full matrix in the design book).

Component DYNASIM This project
Historical earnings Lifetime histories from survey plus linked administrative inputs PSID 1968–2022 panel; gate 1 passed (13th candidate)
Family structure Marriage, divorce, and family in the annual simulation Gates 2, 2b, 2c passed across three tranches
Disability Documented health and work-limitation modules Gate m4 passed, anchored to SSA administrative tables
Base population SIPP-based starting sample Certified CPS synthetic frame; transport gate w1 passed
Forward projection Annual updating and alignment Temporal-holdout gate m6 passed; forward gate pre-lock
Social Security rule engine Rule-based, code not public Open PolicyEngine-US engine plus AIME/PIA oracle (built, not gated)
Claiming behavior Retirement-and-timing model Reduced-form claim-age distribution (built, reported, not gated)
Auxiliary benefits Spouse and survivor analysis in natural domain Spousal, survivor, widow incl. dual-entitlement logic (built, not gated)
Wealth, assets, and LTSS Major documented strength Design-only — a later extension track

Timeline forecast

From the pre-registered timeline ledger, entry 22 (registered 2026-09-03). Dates are point-in-time forecasts, not commitments; the ledger records every revision with its reasons.

Milestone p50 p80
Milestone A — revision-22 ratification (A20 framework operative) 2026-09-10 2026-09-17
Milestone B — corrected-earnings successor authority and publication (the resolution event) 2026-10-21 2026-12-04
  • Milestone A: Not gated by the Q5 program; requires the purpose census (3,230 ranks remaining at registration), the A4 evidence freeze and the C20 ceremony chain.
  • Milestone B: Gated by the Q5 program (45/257 documents, 20.2% of required atoms at registration; two document classes unstarted) and completion of the purpose census.

Recent milestones

  • 2026-09-09 — Morning. Max’s routing ruling arrived through his chief-of-staff session at half past four yesterday afternoon and is now in force: no Opus, Sonnet or Haiku turn on any Claude account that still has Fable headroom, because those turns draw the account’s shared weekly bucket and four accounts were capped that way before their Fable was touched. The orchestrator switched to Fable 5.1 in the same minute and recorded the ruling on the board; every dispatch since has been Fable on a Claude lane or Astra on a Codex lane. The night was a quiet hold on mains power with nothing in flight and the three ruling packages unchanged. A-3 shadow pass 6 was the day’s first decision, and it was run the way the pass-5 adjudication said passes must be run. The seat prepared the worktree and the brief before dawn, with a deltas section carrying the pass-5 acceptance, the pin discipline, the store-write proof, the staging rule, the four failure classes and Max’s ruling. At thirteen past six it asked the orchestrator for a Fable slot on the one Claude lane with headroom; the orchestrator read the lane’s state at that moment rather than an hour earlier, content-asserted the brief and the worktree from bytes, named a foreign run already on the lane, and granted. The seat dry-ran the dispatch twice first, which surfaced a trap the record now calls D15: the plain Fable form of the dispatcher resolves to a read-only sandbox that would have censored the store write on a host that could reach the store, exactly as the Codex sandbox did in pass 5. Store capability means the host and the sandbox mode, and the dry-run is now a required step. The pass landed in nineteen minutes with four commits, inside its thirty-minute bound, and is accepted. It took its own coordinate from the live board, proved that the seventy-two events shadowed through pass 5b reproduce exactly, and then shadowed forty-one new ones: thirty-four harvests, four repairs and three supersessions, among them the orchestrator’s own rotation, the kit’s r10 install and the ruling that voided the old standing order. One hundred and twenty-nine A-2 executions ran with no typed exception and no governing discrepancy; custody measured eighty-six source occurrences and copied none, because no new source has a row in the recorded table. The cumulative count is one hundred and thirteen events across four classes, and H2 remains unreachable before the evening of September 12, when the seven-day floor is met. The lane’s first report claimed it had overrun the bound; it had estimated elapsed time from its own turns instead of reading the clock, caught the error against the commit timestamps, and corrected the report in a fourth commit that touched nothing else. None of yesterday’s four failure classes occurred. The record gained a fifth class an hour later. The farness lane’s five-hour window, shared by the pass, the foreign run and the orchestrator’s own turns, hit its limit as the orchestrator received the acceptance relay, and its seven o’clock heartbeat failed the same way. It resumes automatically at eleven. A limit modal is not a death, so the seat neither rotated it nor spent credits; it holds, runs the orchestrator’s read-only safety instrument on the hour in the meantime, with every axis unchanged, and will relay the acceptance again once the orchestrator is turning. Still with Max: the three ruling packages for gate 3, mortality, and claiming and PIA, and the rotation of the exposed policyengine.org token.
  • 2026-09-08 — Afternoon. The claiming and PIA gates closed the lock discipline to a ruling package in one day. The adversarial referee round on the bound thresholds returned two NOT LOCK-READY verdicts with concrete findings: a mislabelled reading of one rule, a single-rule pass under one grammar, an incomplete flip plan, a footer defect, and a run of contract items. One fixes sitting answered every finding without moving a floor number, and an independent refute-first verification confirmed each fix from bytes, corrected one count in the fixes report from five to three, and settled the flip-time disposition. The ruling package went to Max at 11:16 with seven claiming decisions, five PIA decisions, and the reading of the rule-count definition; the three packages now with him, gate 3, mortality, and claiming and PIA, are the campaign’s critical path. Flip briefs for all three are templated and take nothing but his written words. The C20 kit’s r10 repair passed its review on a scratch clone, with the receipt gate exercised and the fallback unused, and was installed on the orchestrator’s concurrence with one standing requirement: renders made under earlier kit versions must be re-rendered, and any clone used for verification must be a same-filesystem wholesale-object clone, because only the S-prime object is unreferenced and a pack-transfer clone drops it. That custody correction reversed a premise both seats had carried for a day. A-3 shadow pass 5 was dispatched to Astra when no Opus lane answered its clock, and the pass was censored: the Codex sandbox cannot reach the shadow store, so the copy step could not run. The plan owner adjudicated the dispatch as its own error and fixed the rule: an A-3 pass runs only on a host whose writable roots include the store, and a clock with no such host slips and records the slip. Pass 5b completed the pass on a Claude host at the same coordinate. Its lane finished the substance, then died on a DNS error before committing, and a continuation verified the salvage byte for byte and committed it. Accepted: all twenty-nine events shadowed, ninety-three A-2 executions, zero governing discrepancies, cumulative seventy-two events complete; H2 stays not reachable before the evening of September 12. The machine spent the afternoon on battery. No adapter was detected until shortly after three, the lid closed at 14:41 with two percent left and the box slept five minutes, and nothing was lost: everything was committed, and the seat and the orchestrator agreed beforehand that wall-clock silence on battery is uninformative, so no stall or death call may rest on time alone. The unpushed A-3 shadow branch now has a second on-disk copy as a git bundle, proved to restore by reconstruction, without breaching the shadow-only rule. The orchestrator wrote a complete handover line at 14:06 at fourteen percent and, once power returned, asked to be rotated deliberately at eighty-five percent of its transcript trigger rather than die mid-task; its successor proved the safety baseline on arrival. The rotation surfaced a hygiene finding: the launcher passed the lane token on the tmux command line, and because tmux keeps the first client’s arguments as its server’s, the shared server has shown a token in the process list since September 6. Both launchers now fetch the token inside the pane, the successor’s launch added nothing to any command line, a sweep of every running process found no other credential, and the rotation of the exposed token is with Max. The shift’s failures are filed in four classes, Fable weekly caps, a DNS outage, machine sleep, and a limit modal that read as death, and only the first is capacity. The box went back to battery at 15:57, which is why this refresh ran at four rather than eight.
  • 2026-09-08 — Morning. The claiming and PIA gates went through three steps of the lock discipline overnight. Floors v2 answered both adversarial referees in one commit with every alternative emitted rather than chosen: three revision-sd grammars at three K values, a K-by-rounding sweep, settle-age strata with horizon-aware pricing, an OLS window sweep that shows the knife-edge cell is a property of the scanned window grid, and a stated decimal rounding mode with its one exact tie and thirteen-cell knife-edge class. No drafted tolerance or partition count moved; six open decisions are filed for the ruling, and a seventh packet correction records that the structural demotion of the knife-edge cell is unavailable as written because the transition it cites begins a year later. An independent verification then confirmed every fix from bytes under its own code, adjudicated the one referee number that had not reproduced in the statistical referee’s favour, and found one new low finding, a fourth committed channel carrying the held-out rows. The threshold-binding sitting followed: two gate-schema artifacts derived from the floors by path with their digests pinned, every filed alternative’s partition and tolerances carried beside the draft so a ruling changes a pointer rather than a number, LOCKED-HOT bindings that recompute all forty-two tolerances from the floor artifact, and both DRAFT gate blocks carried inside the artifacts as strings with a placeholder for every ruling. Under the definition’s letter the PIA partition awards exactly one rule all three conditions. The seat reproduced the tests. Then the capacity ran out. Every Fable lane hit its weekly limit by three in the morning, and the adversarial referee round on the bound thresholds died on the cap. A retry loop burned eighteen empty dispatches before the orchestrator measured why: a one-line probe answers on a lane whose per-model budget a full brief exhausts. The dispatcher now demotes a lane after a failed dispatch, and the two referees are running on the one Fable lane still open. The seat itself stopped a third time at 03:14: its own scheduled wake did not fire, and the app paused the session two hours later; the cause is undetermined. A one-shot revival at 05:31 repaired the loop and relaunched the seat on Opus as an interim, which found Astra back, dispatched the kit’s r9 two-engine gate, and then hit its own weekly cap; Max resumed the Fable seat at 08:12. The r9 gate failed the candidate with concrete repairs, and an r10 repair draft is running on Astra with the orchestrator’s concurrence step retained. The orchestrator hit its weekly cap the evening before and was relaunched at midnight on a fresh lane by a daemon armed for it. A-3 shadow pass 5 runs on Astra from a pre-created worktree because no Opus lane answered its clock. Gate 3 and the mortality gate remain LOCK-READY and wait on Max’s rulings; H2 stays not yet reachable before September 12.
  • 2026-09-07 — Evening. Gate 3 is ratified LOCK-READY. The third independent verification returned VERIFIED with 24,979 non-string leaves identical across the fix and the two ratifying referees, one statistical and one contract, both returned RATIFY LOCK-READY and agreed on all four decisions: cure A for the description, rho at 0.75, lock the 43-pair tranche under exactly its own description, and the candidate-side module change as a registration precondition. The six typed corrections they conditioned on are applied report-only and verified. The one thing left is the ruling itself, which is Max’s; the flip commit is drafted so it dispatches the moment he answers, with the pre-lock guards inverted and nothing rebuilt. The mortality gate moved through the whole discipline in one afternoon. Floors v3 answered the pre-registration packet’s three data-side blockers from bytes, the weight universe, the death-ascertainment convention and the censoring convention, with the floor rebuilt on the declared universe beside the all-window rebuild. Two adversarial referees followed: the statistical one said LOCK-READY with three findings, the contract one said NOT LOCK-READY narrowly on the record, chiefly that the tests froze the live gates.yaml so the gate block’s own commit would have turned them red. Floors v4 fixed both lists in one commit with no floor number moving, an independent verification returned VERIFIED, and the threshold-binding sitting is running now: the gate-schema artifact, the anchor’s operating characteristic, the draw stream, the teeth table, and a draft gate block carried inside the artifact by digest. The seat itself stopped for eight hours overnight, and the record says why: a completion notice pushed into an idle seat produced no turn, and the app paused the seat fifteen minutes later. The fix has three layers, a self-wake under the pause, a launchd daemon that resumes the session when no process exists, and lane chains that dispatch their own follow-ons. Two more stop mechanisms were found in daylight: the Mac slept on battery until Max plugged it in, now held awake by caffeinate, and every Claude lane hit its weekly limit by mid-morning, leaving one session window as the day’s only capacity. Subfleet’s non-build tasks were also found to run without a shell by default; the referees were re-run with a writable sandbox. A-3 shadow pass 4 accepted on recomputed bytes: fifteen new events, forty-three cumulative, zero disagreeing receipts; H2 stays not yet reachable before September 12.
  • 2026-09-07 — Morning. The ceremony seat stopped for eight hours overnight, and the record says exactly why: at 21:40 the closing sitting’s completion notice was pushed into the seat’s inbox and produced no turn, and at 21:49 the desktop app paused the idle seat after its fifteen-minute timer; nothing ran until 05:43. Two mechanisms, both read from the notify code and the app’s own log, neither a usage limit. The fix has three layers: the seat now wakes itself every thirteen minutes, under the app’s pause; a launchd daemon resumes the session as a one-shot turn whenever no live process exists and hourly as a heartbeat, on a lane it probes first (the top-ranked lane answered ‘weekly limit hit’ in the test); and the lane chains dispatch their own follow-ons, so a landing no longer waits for the seat. Gate 3. The fourth sitting closed the v3 report (332 KB, 33 of 33 digests matching disk, 52 tests). The second independent verification returned NOT VERIFIED with every computed number reproduced by builder-free code, 5,940 of 5,940 panel deviations, 43 of 43 gated rows and 65 of 65 demotions among them, and three blockers all in typed sentences: two false self-descriptions in the committed artifact, the report printing the superseded fragment texts, and a stale test-tier manifest. Floors v4 answered exactly those in one commit (72069f6, five files); the third verification is running and the two ratifying referees are chained behind it. A-3 shadow pass 4 accepted from recomputed bytes: fifteen new events, forty-three cumulative, sixty-two receipts with zero disagreeing, one recorded retry that exposed a pin discipline (pin your own contribution, never the ledger’s totals). H2 stays not yet reachable; the seven-day floor alone binds, earliest 2026-09-12 20:16 EDT. Lane capacity is now the binding constraint rather than the seat: the pass-4 dispatch died on three hard-limited lanes in a row before a fourth carried it, and one non-reserved Opus lane was open at 06:47.
  • 2026-09-06 — Evening. The capability track ran all afternoon on separate lanes beside the correction campaign, and eight shadow branches now sit unpushed and PR-ready, every one with the gate contract byte-identical and its own reproduction tests reproduced by the plan owner: gate-3 retrodiction floors (v1, then v2 answering two adversarial referees, then an independent verification that closed the referees’ blockers by recomputation and opened seven new ones, with v3 in progress); mortality floors v2 (a hundred seeds, the estimand restated as the 1997-onward window); the claiming publication floor and the benefit-oracle rule-coverage artifact; the three registered battery families published with every gate-1 run (full suite 5,705 passed); the report-only entrant schedule with the Trustees net-immigration table captured and hash-pinned; and revision 5 of the immigration design, folding a sixteen-decision referee round. What the lock discipline found on gate 3 is the point of having it: the claiming module’s default reference was the gate’s own resolution table, a textbook drift rule passed 63 percent of the draft surface, no pass rule was stated, and, after v2 fixed all three, an unregistered quadratic-trend rule still passed 23 of 88 gated pairs; the gated set shrinks with each cycle and that is the correct direction. The immigration path has a measured fork for Max: the certified-frame donor is 1,155 rows in 562 households with an effective sample of 82, too thin to floor a gate, so a gated module needs the ACS donor or a coarser surface, while the report-only schedule can ship. Two decisions were re-priced by measurement. The Q5 package: an adversarial pass over the revision candidate’s fourteen folded ranges found nine defects, three material, so the memo’s own rule flips the recommendation to shipping the package of record with the errata and the line-1188 retraction beside it. The ceremony: an attempt candidate without the disputed precommit exception passes the pinned battery at both states, the oracle and the kit’s freeze gate, so ruling (i)’s premise does not hold and the fresh round waits on kit r9’s gate, the R12/R31 ruling and Astra’s return on 12 September. A-3 shadow pass 3 accepted (28 cumulative events; the twelve-event H2 condition met; the seven-day floor alone remains). The seat itself was revived three times today by subfleet onto lane hosts and retired its own duplicates twice; the host ruling stays open.
  • 2026-09-06 — Midday. Max asked for the forecast to DynaSim-class capability and what shortens it, and two lane investigations answered from the record rather than from memory. Capability, as the gate ledger states it: eight pre-registered gates, seven locked, every locked gate with a registered first pass (earnings histories 7 July; marital and fertility 9 July; household composition; marriage x earnings; disability; CPS-frame transport; the temporal-holdout projection engine 24 July), 48 one-shot candidate runs on the record. Only gate 3, forward projection, has never been locked and has no thresholds. Against DynaSim the deficits are: no forward mode past 2022; the model’s earnings are not covered earnings and there is no SSA calibration or OACT alignment (the covered-earnings correction exists to fix the first); no immigration, education or hours modules; mortality floored but not certifiable on the current surface; the benefit oracle and claiming module built and SSA-exact but ungated. Forecast, measured: capability commits over the last six weeks are zero of 140 (57% evidence and ceremony, 19% dashboard), and scored-run cadence is zero since 25 July after 77 artifacts in the 20-day July sprint; no date is derivable from a rate of zero. The only dated plan, the 16 July claim of DynaSim parity in four to seven weeks, expired between 13 August and 3 September; the full-parity milestone (immigration and alignment) is open and past its p80. The binding mechanism is gate 3’s tier-2 arm, which resolves against future annual SSA publications, so its clock cannot start until thresholds are locked; its retrodiction arm needs no publication and is unused. Under redirected capacity the five already-floored links are a 6 to 14 lane-week program; a scored Stage-3 exit is not before Q1 2027. The two top levers cost no lane-weeks because they are decisions: lock gate 3 on the retrodiction arm first, and rule on the graph-port plan’s never-ratified D3 and D6 rules that sequence the benefit gate behind the correction’s publication (p50 21 October). In motion since this morning: the gate-3 retrodiction floor build, the three registered-but-unpublished battery families, pre-registration packets for the claiming, benefit-oracle and mortality gates, and an immigration/entrant design memo, all on separate lanes beside the correction campaign. Also today: A-3 shadow pass 2 accepted (eleven post-H events, the A-2 contract executed over post-H bytes with the plan owner’s per-source custody addendum; H2 not before 12 September); and the decision package’s line 1188, which asserts that no report among the 29 declares an agent count, is measured false (at least seven declare, by file and line) and needs a retraction rather than the errata’s one-word fix, which re-prices the revise-versus-errata decision. The morning’s outage attribution to a usage limit was retracted: the 22:57 kill was the desktop app’s own auto-update; the 23:06 ‘death’ was a one-shot revive exiting normally.
  • 2026-09-06 — Morning after the halted ceremony. Nothing ceremony-related moved overnight and nothing will until three things are ruled on: how the scratch-state battery requirement and candidate-HEAD authentication should coexist (both referees rejected the re-freeze’s reconciliation as an unauthorized precommit exception), whether the two substituted sentences are restored or re-authored, and a kit round that deposits durable evidence of scratch cleanliness; the Codex home that runs the ceremony’s Astra steps is at its weekly cap until 12 September, so no re-run is possible before then in any case. Two corrections to what the decision package tells Max were settled from bytes: the seven-sealed-document itemisation omits one document (a defect the errata already records as its first entry and that four instruments had re-derived independently without any of them updating the package), and the errata’s own Part 2 summary says four decisions change position where its tables and enumeration show nine and eight — its detail is right, its headline is not. The structural question that follows is Max’s: whether the package gets a revision folding the errata’s twenty-seven defects, or ships as-is with the errata beside it. Graph side, all shadow-only and never pushed: plan revision 12 (two acceptance scopes for A-1, a per-class staged rule, a kernel totality contract, and a live-scope requirement for the closing census) was agreed by the peer gate with no findings and merged; the cohort was frozen at an immutable board coordinate with a 4,262-entry content-hashed manifest (5.55 GB across the seven declared source classes, every digest recomputed), the A4 storage record was filled from existing local settings with gated PSID documentation ingested by identity only because the host’s disk is not encrypted, and the H2 stopping rule was frozen before any A-3 run (earliest H2 the evening of 12 September). A-1 accepted at LIVE scope on run 4’s head 3bb53ad: an independent refute-first pass reproduced all eleven clause verdicts with nothing refuted and seven report-level qualifications recorded; run 4’s third declaration rule partitioned the 29 staged-report copies 10/11/8 exactly as predicted; the plan owner reproduced its 212-test suite. Run 4’s delta was accepted on reading and reproduction, not on a second adversarial pass. A-3 pass 1 landed and was accepted as one bounded pass (run 20260906-062442, 19 minutes): 2 post-H harvest events shadowed identity-only with latency recorded, zero discrepancies against bytes, H2 not yet reachable (earliest 2026-09-12). Nothing was copied into the shadow store: the plan’s A4 clause requires the custody checks to be recorded per source before any copy, and the recorded table closed at H, so the pass refused every post-H copy and reported the six sources by identity. The plan owner records the post-H custody rows as a dated addendum before pass 2, which then copies the admissible sources and runs the A-2 contract over them; no plan amendment is needed for custody. Two frozen-enumerator defects (a void pattern without a negation guard; a harvest stated in words that only a creation-receipt sweep caught) are recorded for the next revision. Passes continue daily until H2. Purpose stays at 100.00%; Q5 at 63/257; the twelve decisions remain with Max.
  • 2026-09-05 — Late evening: the Amendment 20 closure ceremony ran to its referee step and stopped there. After the re-frozen candidate C′ (820da19) passed the scratch transition S′ (battery 220/220), the executed-transition receipt (kit gate ready 11 of 11) and a first-add on the ceremony branch, both independent closure referees withheld ratification and wrote no verdict. Both rest on the same two blocking grounds: the re-freeze’s test change, which made the battery hold at the scratch state, reads as a precommit index-authentication exception the parent design does not authorize; and the repository holds no independent evidence that the scratch worktree was clean at S′, a gap in the ceremony’s own evidence design. Both referees also found the candidate’s replacements for the parent’s status-meaning and exact-failure-control sentences not proposition-equivalent; one treats that as blocking, the other does not. Closure and the revision-22 repin were therefore not attempted, the ratification pull request stays unmerged, and the next candidate needs a design-level decision on how the scratch-state battery and candidate-HEAD authentication coexist, a kit round that deposits durable cleanliness evidence, and a ruling on the two sentences. Every step is recorded with digests in the ceremony record. Graph side: the A-1/A-2 build was reviewed clause by clause (accepted at fixture scope; the plan’s live source model not yet met), a shadow child fixed the review’s concrete findings and measured the kernels’ totality (123 tests; three inputs that crash rather than fail typed), and plan revision 12 is drafted with a kernel totality contract and is in its peer gate. Purpose stays at 100.00%; Q5 at 63/257; the twelve decisions remain with Max.
  • 2026-09-05 — Evening: the Amendment 20 A4 freeze was re-frozen after its first candidate failed the pinned battery at the scratch state (a test the freeze itself extended validated the working-tree registry as production whenever ratification was flagged ready, so it could not hold at any scratch state). The re-frozen candidate C′ (820da19, sole parent 86e6994) differs from the first only in that test and its one capture row; the §34 semantic digest is unchanged and was reproduced with the parent commit’s own projection code; the candidate battery passed 220/220 and the scratch-state battery is the remaining gate. Scratch transition S′: dispatched to Astra at 18:01 EDT after the re-freeze report showed readiness 31/31 and both batteries 220/220. Read-only review of the candidate’s prose found that the freeze removed seven sentences from the parent (the three permitting sentences already known, plus a meaning clause, a blanket prohibition on authority selection before activation, and a rule that a proven residue controls) and added at least twelve law-shaped sentences with no parent counterpart, one of which rules on how the campaign’s own board may be read; all reproduced from bytes by a second party and carried to the two closure referees as questions, not rulings. Kit round r8b installed under the two-engine gate: referees now run a whole-document sentence diff in both directions with stated splitter parameters, source-byte probes, and a guard that the withheld-ranks sentence belongs to the missing-reason arm’s occurrence census, not the purpose ledger (which stays at 100.00%). Two figures in the decision package were corrected: the MEASURED-family population is 79 occurrence lines, not 78 (one bracket opens with an en dash that a same-line scan cannot see; four parties had run the same scan), and the tag-id unit is 126 under its author’s own rule (122 numbered items by five independent methods plus four verified §4/§5 items), not 127. The errata’s own ‘largest unswept surface’, a truncation ending mid-clause somewhere in the decision package’s 185 quoted spans, was swept exhaustively: every span ruled (167 complete, 9 truncated at a boundary, 8 altered, 1 not locatable) and zero mid-clause truncations, with the population re-derived blind and all nine control loci recovered. The MEASURED stratum of the decision package was then opened source by source with no wording trigger (78 of 79 loci, 35 of 35 cited files): the source-fidelity defect rate is 11 of 78 (14.1%), so the earlier 13.9% floor is now a bound, and the wording trigger it replaced had found only 3 of those 11; the union of loci carrying a defect on any axis is 15 of 79 (19.0%), stated as a union of two passes rather than one measurement. Both surfaces the errata named as unswept are closed. Graph side: the Plan A shadow plan reached revision 11 through three peer gates and merged to master; its first build slice is running on a Claude lane (Opus) since 17:08 EDT with its first commit (7515f9b, the q5graph package and its test suite) at 17:52, on a Claude lane under the agreed resource policy with no escalation. Q5 stands at 63/257; the twelve decisions remain with Max.
  • 2026-09-05 — The purpose census is COMPLETE: 20,815 of 20,815 prompts adopted (100.00%) at the terminal fold, ledger v1.97 (ebe3d2b9…). Blocks 7–10 of the final tranche folded through the day (v1.93–v1.97), every audit CALIBRATED; block 8 was folded in two parts after an orchestrator scope error (the brief was cut from the per-section table, leaving 64 ranks unadjudicated until an 8b completion pass) and block 9 closed the 2021 questionnaire entirely. A standing correction now requires every lane to recompute each digest it emits from the named file and run its own 64-hex self-check, after a 63-character digest was copied forward from a dispatch brief. A compile-only decision package (twelve decisions the package orders by how much work each blocks, against 63 sealed, 17 held and 1 withdrawn of the 81 annotated documents, although an errata compile finds that order does not consistently follow its own criterion) is staged for ruling; it lists five items it says the commissioning brief missed, the sharpest being that the worktree the lanes run in is two ratified design revisions behind master (19 vs 21), although the slot-catalog section the top decision turns on is byte-identical at both. Two figures already in front of the ruling were corrected: the atomisation coverage gap is 331 of 447 enumerated Social Security, pension, annuity and retirement items in no atom (74.05%, not the 72.7% the census printed and its replication reproduced without recomputing), and the corpus’s page and frame digests are reproducible under the pinned Poppler pdftotext 26.04.0, which the design, the inventory code and the stage-1 replay artifact all name and the code enforces (cross-host determinism untested) (a Chronicle cross-check with pypdf 6.11.0 reproduced 0 of 101 pages, the pinned recipe 101 of 101); the final-tranche pass reports did not restate the build until the block-10 audit did. Q5 stands at 63/257, which is 77.8% of the 81 annotated questionnaires; the remaining 176 files in the fixed domain are setups, codebooks and raw data, not instruments. The open questions are recorded for ruling; the compiler ruled on none of them.
  • 2026-09-05 — Overnight the purpose census reaches 94.88% (final-tranche block 6 folded as ledger v1.92, the fourth consecutive CALIBRATED audit and the first to carry a correction), and the sixteen-document 1968–1982 family back-catalogue is fully annotated and reconciled: six documents sealed, one seal withdrawn on adversarial review, nine held pending rulings (Q5 63/257, which is 77.8% of the 81 annotated questionnaires — the 257 denominator is the fixed source-documentation domain, 43 of which can never seal; a counter reconciliation confirmed the figure across three instruments). A compile-only dossier found the design fixes seven sentinel identifiers (four job-slot, three component-slot) in a section outside every span the briefs had pinned, overturning the campaign’s three-sentinel reading and its bar on the seventh literal; a supersession audit reduced whether the tier-1 minting bar is in force to a single unresolved sentence about Amendment 13’s recording act; and a full-corpus atomisation census, replicated by an independent method, found that 331 of 447 enumerated questionnaire items naming Social Security, pensions, annuities or retirement lie in no atom of the sealed evidence base (66 of 81 documents). Materialization closed with 83 tables of record on disk and exactly two documents provably unrecoverable. All open questions are recorded for ruling, not resolved locally.
  • 2026-09-04 — Final-tranche blocks 4 and 5 close (ledger v1.91; purpose reaches 93.13%), both dense audits CALIBRATED at 100% on full 273-row replays; 1,431 ranks remain across blocks 6–10. Q5 seals q2009, fam2013, q2019 and fam1968 (58/257); fam1968 is the first of a sixteen-document back-catalogue of 1968–1982 family volumes found never briefed and now opened. Five reconciled documents are held and q2023 stays unsealed on the campaign’s first ‘Q5 status does not pass’ verdict, pending five law rulings raised for decision rather than resolved locally: the identifier-minting remedy under §26.5.1, two evidence-integrity classes (digest-only and corrupt sealed payloads), the H-object scope of the pinned law extract, and a minting-clause precondition naming a field that nine artifacts lack. A prepare-only materialization sweep of all 69 reconcile reports writes every reproducible table of record to disk as a first-class artifact with provenance marked decoded or reconstructed (64 tables at last count), and recovery runs prove two of the three corrupt payloads (1995q, 1994f) reproducible from surviving emissions.
  • 2026-09-04 — Final-tranche blocks 1–3 close (ledger v1.89; purpose reaches 90.50%): block 1 on its regeneration sweep, block 2 CALIBRATED at 99.66% on a one-row manifest, block 3 CALIBRATED at 100% with an empty contradiction set. Q5 seals 2003q, 2005f, 2005q, 2007f, 2007q and 2009f (54/257), the last five on Opus reconciliations; the first Opus/Opus pairs measure common-mode rates of 0.4%, 1.8%, 0.0% and 2.9%. The C20 ceremony-kit correction reaches its fifth round as a Sol-written change that an independent Opus review ACCEPTS, clearing the last precondition on the A4 evidence freeze other than the census itself. Six Claude accounts run Opus lanes in parallel after subfleet’s false month-long parks are found and cleared.
  • 2026-09-03 — The E5 first tranche completes: block 8 closes on its audit’s closed 16-row manifest (ledger v1.84; purpose reaches 85.94%). The final tranche R17889–R20815 is planned as the entire remainder — 2,927 ranks across eight documents from 2013 to 2023 in ten blocks (v1.85) — and its first block passes 348/348 and is audited (v1.86) with a 59-row regeneration manifest. Q5 seals 2001f, 2003f and 2001q (48/257), the latter two on Claude Opus reconciliations in 42–43 minutes against a one-to-three-hour Sol baseline.
  • 2026-09-03 — E5 blocks 6–7 close (ledger v1.83; purpose reaches 84.48%) and the tranche’s last block, block 8 to R17888, is in flight. Q5 seals 1999f and 1999q — the latter the campaign’s largest table of record at 8,498 rows — reaching 45/257. An output-order mandate (full serialization first, no placeholders) is adopted as standing process law for large deliverables after a truncated sweep had to be re-run.
  • 2026-09-02 — E5 blocks 3–5 close (block 3 CALIBRATED at 100% — the third perfect purpose block); purpose reaches 81.78%. Q5 seals 1997q after a dual-serialization cross-recovery repaired a damaged lane print without a re-run (43/257).
  • 2026-09-02 — Purpose census enters the E5 era (2003–2011 questionnaires; tranche R15625–R17888 pinned) and reaches 77.53% overnight; Q5 seals 1996q and 1997f (42/257).
  • 2026-09-01 — The A3 classification arm COMPLETES: all 41,103 queue rows final-adopted (49,722 adjudicated ranks across the A1/A2/A3 programs, zero predicate defects campaign-wide). Purpose census reaches 71.00% with its last in-scope document (q2001) in the verification cycle; Q5 reaches 38/257.
  • 2026-08-30 — Evening sprint: purpose reaches 60.97% (blocks 5–7 closed; q94–q95 complete; the q96 opener — E4’s final document — enters audit); A3 reaches 93.4% with its first two audit-clean blocks ever; Q5 seals 1992q (34/257) under a new occurrence-local binding-law ruling.
  • 2026-08-30 — Purpose census reaches 55.61% (q94 B+C closed under ledger v1.44 after the audit reversed the opener’s imported simultaneity frame — the cross-document transfer prohibition holding under pressure); A3 frontier reaches 35,200/41,103 (85.6%).
  • 2026-08-30 — E3 questionnaire era completed and the E4 era opened overnight; f94 complete; purpose census reaches 54.10% contiguous. A3 reaches the 1968-era queue boundary (83.7%). Q5 seals documents 048–049 (33/257).
  • 2026-08-29 — Purpose census crosses 50% of the residual domain (contiguous R1–R10457 = 50.24%); q93 B+C complete.
  • 2026-08-28 — f91 complete; purpose census at 43.11% contiguous; governing ledger reaches v1.26.
  • 2026-08-26 — f89-era tranche closed: contiguous R1–R8552 = 41.09%; every rank of the 1,920-rank tranche went through pass, dense audit, adjudication where audits split, and correction sweep.
  • 2026-08-20 — Timeline-forecast entry 21 registered: scope bifurcation into milestones A and B.

How these numbers are produced

Every block of work moves through the same pipeline before it counts here: a full pass over the block, an independent dense audit, an adjudication step where audits disagree, and a correction sweep whose output is folded into the governing ledger. Q5 documents are annotated by two independent lanes and sealed only by a reconciliation of both. Nothing on this page is self-reported by the lane that produced it.