polymath-society 0.2.5 → 0.2.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -661,7 +661,7 @@ var GIFTS = [
661
661
 
662
662
  // ../../lib/calibration/index.ts
663
663
  var DEFAULT_CALIBRATION = {
664
- version: "2026-07-12.2",
664
+ version: "2026-07-14.2",
665
665
  // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
666
666
  // src/grade/criteria.ts). A unit test fails loud when the rubric changes
667
667
  // without this being updated (and re-pushed to the server).
@@ -750,7 +750,7 @@ var DEFAULT_CALIBRATION = {
750
750
  }
751
751
  ],
752
752
  codingTiers: {
753
- push: { median: 300, sigma: 1.121, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail shape from its actions-per-prompt distribution. AVERAGE over substantial days, cumulative not spiky." },
753
+ push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
754
754
  focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
755
755
  orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
756
756
  },
@@ -783,12 +783,18 @@ var DEFAULT_CALIBRATION = {
783
783
  line: "the Claude Code lead runs 10\u201315 sessions at once; his #1 tip is 3\u20135 git worktrees in parallel"
784
784
  },
785
785
  throughput: {
786
- // Top-0.1% words/day reference lines, anchored to the throughput ladder
787
- // (THROUGHPUT_LADDER in throughput.ts): score 10 "world-class sustained" =
788
- // 9k/day, score 11 "superhuman ceiling" = 12k/day. Calibrated to the person's
789
- // own anchors, not a wild guess but still a target, not a measured population.
790
- topAvgWordsPerDay: 9e3,
791
- topPeakWordsPerDay: 12e3,
786
+ // Two DIFFERENT kinds of line (2026-07-14.2): topAvg is a PERCENTILE
787
+ // claim — what a 1-in-1,000 user AVERAGES over substantial days (5k/day:
788
+ // near-daily heavy dictation; a measured top dictating power user
789
+ // averages ~3.9k). topPeak is NOT a percentileit is the CEILING
790
+ // reference, the wildest single days on record (~10k: a 14-16h
791
+ // launch-day marathon), rendered as "wildest single days", never
792
+ // "top X% peak". An avg-percentile and a peak-percentile can't both be
793
+ // shown at a same-ish value (avg≈peak reads broken — user call), and a
794
+ // peak PERCENTILE line is unknowable anyway; the record framing is
795
+ // honest and keeps the visual gap.
796
+ topAvgWordsPerDay: 5e3,
797
+ topPeakWordsPerDay: 1e4,
792
798
  benchLabel: "top 0.1%",
793
799
  // No published "words typed/day" for heavy users — the real story is OUTPUT.
794
800
  loopsHeadline: ">1,000,000 lines of Rust at 99.8% tests passing",
@@ -15635,7 +15635,7 @@ var ENGINE_DATA = process.env.POLYMATH_DATA_DIR ? path10.resolve(process.env.POL
15635
15635
 
15636
15636
  // ../../lib/calibration/index.ts
15637
15637
  var DEFAULT_CALIBRATION = {
15638
- version: "2026-07-12.2",
15638
+ version: "2026-07-14.2",
15639
15639
  // = RUBRIC_FINGERPRINT of the shipped npm rubric (packages/coding-analyzer/
15640
15640
  // src/grade/criteria.ts). A unit test fails loud when the rubric changes
15641
15641
  // without this being updated (and re-pushed to the server).
@@ -15724,7 +15724,7 @@ var DEFAULT_CALIBRATION = {
15724
15724
  }
15725
15725
  ],
15726
15726
  codingTiers: {
15727
- push: { median: 300, sigma: 1.121, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail shape from its actions-per-prompt distribution. AVERAGE over substantial days, cumulative not spiky." },
15727
+ push: { median: 300, sigma: 0.91, tag: "estimated \u2014 log-normal fit", source: "Anthropic 2026 Claude Code study (400k sessions): median engaged user ~8 prompts/day at ~35 words \u2014 ~300 directed words/day; tail: a 1-in-1,000 user AVERAGES ~5k directed words/day (near-daily heavy dictation, clearly below the ~10k single-day record). AVERAGE over substantial days, cumulative not spiky." },
15728
15728
  focus: { median: 0.08, sigma: 0.559, tag: "estimated \u2014 proxy-anchored", source: "median: Anthropic 20h/week engaged-user figure implies ~1h/day of true rapid exchange against an 8h working day; ceiling anchored to the ~4h/day deep-work limit (45% share = 1-in-1,000). AVERAGE share, cumulative not spiky." },
15729
15729
  orchestration: { tag: "estimated \u2014 behavior bands", source: "no published concurrency distribution exists; ceiling anchored to the Claude Code lead's documented 10-15 parallel sessions" }
15730
15730
  },
@@ -15757,12 +15757,18 @@ var DEFAULT_CALIBRATION = {
15757
15757
  line: "the Claude Code lead runs 10\u201315 sessions at once; his #1 tip is 3\u20135 git worktrees in parallel"
15758
15758
  },
15759
15759
  throughput: {
15760
- // Top-0.1% words/day reference lines, anchored to the throughput ladder
15761
- // (THROUGHPUT_LADDER in throughput.ts): score 10 "world-class sustained" =
15762
- // 9k/day, score 11 "superhuman ceiling" = 12k/day. Calibrated to the person's
15763
- // own anchors, not a wild guess but still a target, not a measured population.
15764
- topAvgWordsPerDay: 9e3,
15765
- topPeakWordsPerDay: 12e3,
15760
+ // Two DIFFERENT kinds of line (2026-07-14.2): topAvg is a PERCENTILE
15761
+ // claim — what a 1-in-1,000 user AVERAGES over substantial days (5k/day:
15762
+ // near-daily heavy dictation; a measured top dictating power user
15763
+ // averages ~3.9k). topPeak is NOT a percentileit is the CEILING
15764
+ // reference, the wildest single days on record (~10k: a 14-16h
15765
+ // launch-day marathon), rendered as "wildest single days", never
15766
+ // "top X% peak". An avg-percentile and a peak-percentile can't both be
15767
+ // shown at a same-ish value (avg≈peak reads broken — user call), and a
15768
+ // peak PERCENTILE line is unknowable anyway; the record framing is
15769
+ // honest and keeps the visual gap.
15770
+ topAvgWordsPerDay: 5e3,
15771
+ topPeakWordsPerDay: 1e4,
15766
15772
  benchLabel: "top 0.1%",
15767
15773
  // No published "words typed/day" for heavy users — the real story is OUTPUT.
15768
15774
  loopsHeadline: ">1,000,000 lines of Rust at 99.8% tests passing",