Conversation
There was a problem hiding this comment.
Pull request overview
Adds backlog row B-0196 to track research on arbitrary-precision numerics in Zeta, framing it as a survey-and-decision item rather than an implementation change. This fits the repo’s backlog substrate by recording a new research question and surfacing it in the generated backlog index.
Changes:
- Adds a new P2 backlog row for BigInteger / BigRational / BigDecimal / BigFloat / SRTP-generic numerics research.
- Defines acceptance criteria, candidate risk sites, and decision branches for follow-up work.
- Regenerates
docs/BACKLOG.mdto include the new row in the P2 index.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 8 comments.
| File | Description |
|---|---|
docs/backlog/P2/B-0196-bigint-and-bignumber-integration-aaron-2026-05-05.md |
New backlog row describing the numerics-integration research scope, baseline, falsifiers, and related backlog links. |
docs/BACKLOG.md |
Generated backlog index updated to list B-0196 under P2. |
AceHack
added a commit
that referenced
this pull request
May 5, 2026
…Gamma + tools/z3->tools/Z3Verify + + bullet + BayesianAggregate Checked correction (#1595 reviewer) Seven reviewer threads addressed: 1. **Bidirectional composes_with** (per tools/backlog/README.md:69-70): added B-0196 to B-0140, B-0156, B-0189, B-0194's composes_with lists. B-0140 + B-0156 had no composes_with field; created. 2. **`+` continuation bullet** at line 139 broke markdownlint MD007 (parsed as nested + list). Changed `+ provides a count` to `AND provides a count`. 3+5. **NormalGamma -> NormalInverseGamma**: BayesianAggregate.fs has NormalInverseGamma, not NormalGamma. Sed-corrected at all occurrences. 4. **BayesianAggregate Checked.* survey baseline**: reviewer correctly noted BayesianAggregate.fs has no Checked.* arithmetic guards (only one doc-comment mention). The original subagent's listing was wrong. Corrected with explicit acknowledgment + note that Bayesian surfaces use double arithmetic without checked guards (different shape from int64- overflow risk; relevant to BigRational candidate, not BigInteger). 6. **Units.fs cross-reference**: was stale at PR time. Now resolves on main since #1590 merged. 7. **tools/z3/ -> tools/Z3Verify/**: corrected dead path reference. Regenerated docs/BACKLOG.md. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…gration -- substrate survey + per-class adoption recommendation (Aaron 2026-05-05) Aaron 2026-05-05 verbatim "oh backlog bigint and other bitnumbers integration" -- surfaced after a reviewer caught int64 overflow in Units.msToNs (PR #1590, fixed via Checked.(*) + 3 overflow tests). The msToNs case is one instance of a broader question this row owns: should Zeta support arbitrary-precision integers (BigInteger), arbitrary-precision rationals (BigRational), arbitrary-precision decimals (BigDecimal), and arbitrary-precision floats (BigFloat / MPFR-style) for cases where int64 / double overflow or precision- loss is in the operating regime? Research + plan row, NOT implementation. Output is a survey + per- numeric-class adoption recommendation; doing any swap that turns out to be warranted is a separate row filed downstream. Five concrete shapes covered: BigInteger (.NET System.Numerics), BigRational (exact probability arithmetic), BigDecimal (financial, LFG context), MPFR-style BigFloat (scientific), and SRTP-based generic numerics (numeric type-class abstraction). Acceptance criteria with falsifiability hooks: - (a) Substrate survey -- baseline numbers in row: ~470 raw int64|float|double mentions in src/Core, 40 Checked.* arithmetic guard sites across 12 files. - (b) Risk-site identification -- 5 starting candidates each with operating regime + falsifier: msToNs, Z-set weight aggregation, IndexedZSet bilinear delta product, BayesianAggregate (BetaBernoulli / NormalGamma / DirichletMultinomial), TimeSeries window arithmetic. - (c) Decision -- per-class adoption vs system-wide generic-numerics refactor vs defer-until-second-overflow-site (carved-sentence default). - (d) F# UoM compatibility note -- BigInteger likely not on the F# spec's allowed-measure-types list; needs WebSearch verification per Otto-364 search-first authority. Carved sentence: *"Pick numerics by where the regime fires; per- class adoption beats a system-wide refactor until the second overflow site lands. The msToNs fix is a Checked-throws guard, not yet a BigInteger argument; the second site is the trigger."* Composes with B-0140 / B-0156 (TS+Bun BigInt sister), B-0189 (Q# Bayesian -- where precision matters), B-0194 (capability dispatcher -- where the bilinear-product overflow lives). P2 (not P1, not P3): higher than P3 because real trigger (PR #1590 reviewer caught overflow) and Aaron explicitly named the row; lower than P1 because the Checked.* guard already shipped is sufficient for correctness, and the BigInteger / BigRational / BigDecimal decision is research-grade, not blocking. Substrate-or-it-didn't-happen (Otto-363): converts Aaron's chat- weather framing into a falsifiable durable research artifact. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…-05-05 same-tick)
Aaron 2026-05-05 added two pieces of context after the initial
subagent draft was committed:
1. **Four-property hodl invariant as binding acceptance test**:
DST-safe + lock-free + scale-free + DBSP-native must all hodl
for any numeric-type candidate. NOT separable -- relaxing any
one lets bugs through (the Prop 3.5 misattribution caught
earlier this session is a worked example). Composes with the
"DST hodls everywhere" lineage at memory/feedback_aaron_class_
discovery_experiment_rodney_razor_on_self_dst_holds_everywhere
_aaron_2026_05_01.md. Per-candidate acceptance scoring against
the conjunction, not score-against-overflow-prevention-alone.
2. **Human anchors** named:
- **Tanner Gooding** (.NET runtime, generic math interfaces) --
primary anchor for SRTP-replacement-for-numeric-typeclass shape
- **Don Syme** (F# language) -- stays the F# anchor for
language-extension work (RFC #1591); numerics-runtime goes
through Gooding
- **Leonid Ryzhyk (@ryzhyk)** (DBSP / Feldera GitHub) --
ENGAGEMENT GATE: only engage IF a real DBSP-compatibility
finding lands (cross-check first; speculative findings don't
reach external engagement -- the Prop 3.5 misattribution was
itself a worked example of why).
Carved-sentence-wise: "BigDecimal-with-rounding may NOT satisfy
DBSP-native (rounding errors don't necessarily compose with
negation)" is one operationally testable per-candidate result.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…Gamma + tools/z3->tools/Z3Verify + + bullet + BayesianAggregate Checked correction (#1595 reviewer) Seven reviewer threads addressed: 1. **Bidirectional composes_with** (per tools/backlog/README.md:69-70): added B-0196 to B-0140, B-0156, B-0189, B-0194's composes_with lists. B-0140 + B-0156 had no composes_with field; created. 2. **`+` continuation bullet** at line 139 broke markdownlint MD007 (parsed as nested + list). Changed `+ provides a count` to `AND provides a count`. 3+5. **NormalGamma -> NormalInverseGamma**: BayesianAggregate.fs has NormalInverseGamma, not NormalGamma. Sed-corrected at all occurrences. 4. **BayesianAggregate Checked.* survey baseline**: reviewer correctly noted BayesianAggregate.fs has no Checked.* arithmetic guards (only one doc-comment mention). The original subagent's listing was wrong. Corrected with explicit acknowledgment + note that Bayesian surfaces use double arithmetic without checked guards (different shape from int64- overflow risk; relevant to BigRational candidate, not BigInteger). 6. **Units.fs cross-reference**: was stale at PR time. Now resolves on main since #1590 merged. 7. **tools/z3/ -> tools/Z3Verify/**: corrected dead path reference. Regenerated docs/BACKLOG.md. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
AceHack
force-pushed
the
backlog-bigint-bignumber-integration-aaron-2026-05-05
branch
from
2083a04 to
e582f38
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e582f38480
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…56/B-0189 (#1595 reviewer follow-up) Three reviewer + CI findings: 1. MD032/blanks-around-lists at line 148: list "- src/Core ..." needed blank line before. Fixed. 2. MD018/no-missing-space-atx at line 412: "#1590 with Checked.(*)..." at line-start parsed as ATX heading. Rephrased "PR #1590" to "PR 1590" in the soft-wrapped paragraph; ATX-collision avoided. 3. last_updated bumps: B-0156 (2026-05-01 -> 2026-05-05) and B-0189 (2026-05-04 -> 2026-05-05) had frontmatter edits adding B-0196 to composes_with but kept stale dates. Bumped per tools/backlog/README.md schema requirement. Regenerated docs/BACKLOG.md. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…earch-required (#1595 reviewer) Reviewer correctly caught: section (b) asserted as fact ("F# UoM is constrained to primitive numeric types per the F# spec; this is itself a substantive finding") what section (d) treats as unverified research. Pre-answering one's own research question is the failure mode the four-property hodl is meant to catch -- the training-data prior was leaking through as if it were verified fact, contradicting Otto-364 search-first authority. Reframed: BigInteger<weight> is a hypothesis to test. The expected training-data answer is the primitive-numeric-types constraint, but that hypothesis must be WebSearch-verified against current upstream documentation before binding. Recent F# versions may have lifted the constraint, or community workarounds (measure-attributed wrapper structs) may exist. Section (d) is the research task that resolves this. The two stale last_updated reviewer threads (PRRT_kwDOSF9kNM5_kaHu on B-0140, PRRT_kwDOSF9kNM5_kaIL on B-0194) are reviewer-snapshot stale -- e582f38 already bumped B-0140 to 2026-05-05, and B-0194 was created at 2026-05-05 from origin. Both currently correct on the branch. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Comment on lines
+106
to
+108
| (`'T : ( + ) ( * ) Zero One`) rather than hard-coded | ||
| to `int64` / `double`. Existing F# patterns: | ||
| `LanguagePrimitives.GenericZero`, |
Comment on lines
+268
to
+274
| adopt SRTP-based numeric type-classes | ||
| (`'T : ( + ) ( * ) Zero One`) so all DBSP | ||
| operators are numeric-type-polymorphic. Default | ||
| if (b) finds 3+ sites AND the SRTP performance | ||
| cost is judged acceptable. Higher scope, higher | ||
| disruption, future-proofs against new | ||
| numeric-type asks. |
| 3. **Scale-free** -- works across scales, small to BigInteger / arbitrary-precision. | ||
| 4. **DBSP-native** -- retraction-aware. Negative weights compose cleanly through the operation; the operation is consistent under signed-delta application. NOT all numeric types satisfy this trivially: BigInteger as weight type works (ring algebra holds); BigDecimal-with-rounding may NOT (rounding errors don't necessarily compose with negation); float arithmetic explicitly doesn't (commutative-but-not-associative under accumulation). | ||
|
|
||
| **Per-candidate acceptance scoring**: each numeric-type candidate (BigInteger, BigRational, BigDecimal, BigFloat) gets scored against all four. Failure on ANY axis means rejection or restriction-to-non-incremental-use. Score-against-the-conjunction is the binding test, not score-against-overflow-prevention-alone. |
Comment on lines
+420
to
+421
| question" -- it was "BigInt and other bignumbers | ||
| integration" as a broader substrate question. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Units.msToNs(PR feat(core/units): F# UoM declarations for Zeta domain (Aaron 2026-05-05) #1590, fixed viaChecked.(*)+ 3 overflow tests).Acceptance criteria (with falsifiability hooks)
int64|float|doublementions insrc/Core, 40Checked.*guard sites across 12 files.Composes with
Why P2 (not P1, not P3)
Checked.*guard already shipped is sufficient for correctness; the BigInteger / BigRational / BigDecimal decision is research-grade, not blocking.Test plan
docs/backlog/P2/B-0196-bigint-and-bignumber-integration-aaron-2026-05-05.mdwith frontmatter conforming totools/backlog/README.mdschema (id, priority, status, title, tier, effort, ask, created, last_updated, depends_on, composes_with, tags).tools/backlog/generate-index.sh --checkreturns clean (docs/BACKLOG.mdregenerated and matches generator output).docs/BACKLOG.mdindex under P2 section.docs/AGENT-BEST-PRACTICES.mdfalsifier discipline.🤖 Generated with Claude Code