Disposition doc for eigentrust pitfalls #8/#9/#10/#16 (observations)

docs/tracking/2026-04-25_eigentrust_pitfalls_observations.md

@@ -0,0 +1,96 @@
+# EigenTrust Pitfalls — Observation Items (8, 9, 10, 16)
+
+_Companion doc to `docs/tracking/2026-04-23_eigentrust_pitfalls.md` (when that
+branch lands). The 2026-04-23 memo enumerated 16 pitfalls hit during the
+EigenTrust implementation. Items 1–7 and 11–15 are language/elaboration
+defects with their own fix PRs. This doc dispositions the four items that are
+**observations** rather than Prologos defects, so no compiler code change is
+required for them._
+
+## Pitfall 8 — Exact-Rat arithmetic slows power iteration by orders of magnitude
+
+**Disposition: not a defect; benchmark-scope guidance.**
+
+The slowdown is intrinsic to exact rational arithmetic: each multiplication
+can grow numerator and denominator by a small factor, so after *k* steps the
+denominators can reach `≈ (10 · max-matrix-denom)^k`. The simplification pass
+runs on every operation but cannot prevent growth when the iterates do not
+share a common denominator. This is fundamental to choosing exact-Rat as the
+numeric domain.
+
+**Action**: continue using exact-Rat for *correctness* tests (golden-output
+testing of small fixtures) and use Posit32 / Float variants for
+*deep-iteration* benchmarks. The PVec+Posit32 EigenTrust variants in
+`benchmarks/comparative/` already follow this split.
+
+If iterate growth ever becomes a Rat-correctness blocker rather than a
+performance one, two paths exist:
+- Common-denominator normalization across the iterate vector after each
+  step (gives bounded growth at the cost of `gcd`-per-step).
+- Switch to a fixed-precision rational representation (rational with a
+  capped denominator).
+
+Neither is justified by current usage; revisit only if a real workload
+demands it.
+
+## Pitfall 9 — Posit32 literals survive nested list/PVec literals correctly
+
+**Disposition: positive observation; no action.**
+
+Where `'[0/1 1/2]` mis-types the Rat `0/1` as Int (pitfall #3, fixed under its
+own PR), the Posit literal forms `'[~0.0 ~0.5]` and `@[~0.0 ~0.5]` both
+preserve `Posit32` correctly. The reason: the `~` literal prefix has no
+bare-Int alias, so the preparser cannot silently re-interpret it as a
+different numeric type.
+
+**Validation**: covered indirectly by the EigenTrust Posit benchmarks
+(`benchmarks/comparative/eigentrust-pvec-posit.prologos`). Worth adding an
+explicit test if pitfall #3's fix introduces any preparse changes that touch
+the numeric-literal path; otherwise no work needed.
+
+## Pitfall 10 — PVec `@[...]` literals preserve element types where `'[...]` does not
+
+**Disposition: subsumed by pitfall #3 fix.**
+
+This is the dual side of pitfall #3. `@[0/1 1/2]` correctly elaborates as
+`PVec Rat`; `'[0/1 1/2]` mis-elaborates the `0/1` to Int. Pitfall #3's PR
+brings the List path in line with the PVec path (or vice-versa, depending on
+the chosen fix shape).
+
+After pitfall #3 lands, both literal forms preserve element type uniformly.
+No separate fix is needed for this pitfall — close as a duplicate of #3.
+
+## Pitfall 16 — EigenTrust wants column-stochastic, not row-stochastic
+
+**Disposition: algorithm/spec clarification, not a Prologos defect.**
+
+The original 2003 paper defines `c_{ij}` as "peer i's normalized trust in
+peer j", giving a row-stochastic `C`. The actual update step is
+`t_{k+1} = C^T · t_k` — i.e. `C^T` is what gets multiplied by `t`, and
+`C^T` is column-stochastic.
+
+The first List+Rat implementation took row-stochastic `C` and computed
+`(C^T · t)` internally via "sum of row-scaled rows" (a transpose-avoiding
+trick). This worked but made the invariant awkward to state, and required
+users to understand the implementation's trick to interpret the input.
+
+**Action taken in the EigenTrust impl**: take the **column-stochastic** matrix
+`M` directly as input. The update is plain matrix-vector multiply
+`y[i] = dot(M[i], t)`. The invariant ("each column of M sums to 1") is
+checkable via `col-stochastic?` in `O(n²)` and the `eigentrust` entry point
+panics if it fails. The "ring fixture" (a column-stochastic permutation
+matrix) is a cleaner slow-iteration benchmark than an asymmetric
+row-stochastic matrix relying on the transpose trick for interesting dynamics.
+
+This is captured in the EigenTrust implementation branch
+(`claude/eigentrust-prologos-implementation-JGszt`) and is a benchmark/algorithm
+shape choice. No Prologos compiler change is required.
+
+## Cross-references
+
+- Pitfall fixes (one PR per): #1, #2, #3 (subsumes #10), #4, #5, #6 (already
+  landed as #4), #7, #11, #12, #13 (#5 PR), #14 (#6 PR), #15.
+- This doc: #8, #9, #10, #16 (observations only).
+
+When the 2026-04-23 pitfalls memo lands on `main`, link this doc from its
+"Status" or "Disposition" section.

racket/prologos/benchmarks/micro/info.rkt

@@ -0,0 +1,12 @@
+#lang info
+
+;; Pre-0 micro-benchmarks for tracks that were implemented after this
+;; file was last updated. `bench-bsp-le-track2.rkt` references the
+;; pre-D.5b TMS API (`tms-write`, `tms-cell-value`, `tms-read`,
+;; `tms-commit`) that no longer exists; since the file is a historical
+;; baseline-measurement artifact (not code that runs in CI or
+;; regression suites), we simply skip it during `raco setup` compilation
+;; so it does not fail the build. To re-enable the benchmark, migrate
+;; its calls to the current `tms-cell` / `atms-write-cell` API in
+;; `atms.rkt` and remove this omit directive.
+(define compile-omit-paths '("bench-bsp-le-track2.rkt"))