feat: disclose RankWeave ranking channel evidence (v2.12.17)

AGENTS.md

@@ -36,6 +36,8 @@ reimplementing them:
   multi-channel score fusion (`weighted_convex_fuse` in
   `reconstruct.py`) and the buyer-facing Rankings port
   (`rankweave_client.py`) -- never invent a fused score or a theta.
+  Rankings disclose Cormack RRF channel contributions from owned
+  rank lists (ADR 0167).
 - [TEPP](https://github.com/ContextualWisdomLab/TEPP)'s published wire
   contract for calibrated measurement (`tepp_client.py`) -- never
   reimplement TEPP's model here.

ARCHITECTURE.md

@@ -122,7 +122,10 @@ flowchart LR
   `rankweave_client.py`'s default transport raises
   `RankWeaveNotAvailable`. `GET /api/rankings` then returns
   `rankweave_not_available` and an empty ranking list. Hidden posts
-  are omitted from every channel. See ADR 0024.
+  are omitted from every channel. Accepted hits include
+  `channel_evidence` computed from owned temporal/lexical ranks
+  (Cormack weighted RRF contribution); RankWeave extra fields are
+  ignored and no theta is invented. See ADR 0024 and ADR 0167.
 
 ## Standards and citations
 

CHANGELOG.d/2.12.17-rankweave-ranking-channel-evidence.md

@@ -0,0 +1,11 @@
+# 2.12.17 — RankWeave ranking channel evidence
+
+Rankings now shows why a fused hit landed.
+
+- `GET /api/rankings` attaches owned-channel evidence: 1-based
+  newest-first and title-overlap ranks plus Cormack RRF contribution
+  `weight / (η + rank)` with η = 60.
+- A channel the post is missing from is omitted. RankWeave extra
+  fields are ignored. No fused score or theta is invented.
+- The Rankings list discloses that evidence under each hit. Copy
+  states this is not a calibrated score. Click still opens the post.

CHANGELOG.md

@@ -25,6 +25,15 @@ All notable changes to this project are documented here. Format follows
   environment, so local OIDC and synthetic-data workflows resolve the same
   pinned dependencies as CI.
 
+## [2.12.17] - 2026-08-24
+
+### Added
+
+- Rankings now names the newest-first and title-overlap ranks that
+  RankWeave fused, with each channel's Cormack RRF contribution.
+  Missing channels stay omitted. This is ranking evidence, not a
+  calibrated score. Click still opens that post.
+
 ## [2.12.15] - 2026-08-24
 
 ### Added

backend/app/main.py

@@ -3123,11 +3123,12 @@ async def read_rankings(
     account: CurrentAccount = Depends(get_current_account),
     pool: asyncpg.Pool = Depends(get_pool),
 ) -> dict[str, Any]:
-    """RankWeave fusion of ABAC-visible posts (ADR 0024).
+    """RankWeave fusion of ABAC-visible posts (ADR 0024 / ADR 0167).
 
     Hidden posts are omitted from every channel. Never invents a fused
-    score or a theta. Fail-closed when RankWeave is disabled or the
-    library is missing.
+    score or a theta. Channel evidence is computed from owned rank
+    lists. Fail-closed when RankWeave is disabled or the library is
+    missing.
     """
     _require_post_read(account)
     async with pool.acquire() as conn:

docs/adr/0024-rankweave-fusion-fail-closed.md

@@ -35,6 +35,11 @@ tables, and does not bind the demo IdP to production Keyverse.
 4. After login, Rankings sits above Calendar. Unavailable copy is
    **Rankings · RankWeave not available**. An accepted hit lists the
    title; click opens that `source_post`.
+5. Accepted hits also disclose owned-channel evidence (ADR 0167):
+   1-based `channel_rank` and Cormack contribution
+   `weight / (η + rank)` for each channel the post actually appears
+   in. Missing channels are omitted. RankWeave extra fields are
+   ignored. Copy states this is not a calibrated score.
 
 ## Consequences
 

docs/adr/0167-rankweave-ranking-channel-evidence.md

@@ -0,0 +1,67 @@
+# ADR 0167 — Disclose RankWeave ranking channel evidence
+
+**Decision status:** Accepted
+**Date:** 2026-08-23
+
+## Context
+
+ADR 0024 already fuses ABAC-visible posts through RankWeave weighted
+reciprocal-rank fusion (Cormack et al., 2009, η = 60; Samuel et al.,
+2025, unequal weights). `GET /api/rankings` returned only `post_id`,
+`post_title`, and 1-based `fused_rank`. A reader could open the hit
+but could not see which owned channel ranked it, or how much that
+rank contributed.
+
+Event Lineage channel evidence (ADR 0124 on #387) explains
+reconstructed parent→child edges from persisted convex-fusion
+scores. Rankings is a different surface: two rank-only channels
+(`temporal`, `lexical`) fused in-process at GET time. There is no
+persisted ranking table and no TEPP theta. RankWeave extra fields
+must not be trusted; a missing channel stays missing.
+
+This ADR does not replace reconstruction fusion, leftover pairs,
+TEPP receipts, Allen interval labels, or Event Lineage isolation
+reasons.
+
+## Decision
+
+1. Compute ranking channel evidence from LineageWeave-owned rank
+   lists, never from RankWeave payload extras. For each fused hit
+   and each channel with a positive weight, take the 1-based rank
+   of that `post_id` in the ordered id list. Skip a channel the
+   post is absent from. Do not invent a rank.
+2. Contribution is Cormack weighted RRF:
+   `weight / (η + rank)` with η = 60. Sort by contribution
+   descending, then `signal_code`. `rank` on the evidence row is
+   that 1-based evidence order. `channel_rank` is the 1-based
+   position in that channel.
+3. Labels: `temporal` = **Newest first**, `lexical` = **Title
+   overlap**. The payload never includes a fused score or a theta.
+4. Rankings lists the evidence under each accepted hit as an
+   accessible sibling list, not hover-only. Copy states that
+   RankWeave fused newest-first and title-overlap ranks, and that
+   this is not a calibrated score. Click still opens that post.
+5. Unavailable RankWeave stays empty (`rankweave_not_available`).
+   Hidden posts remain omitted from every channel.
+
+## Consequences
+
+Buyers can see why a Rankings hit landed without treating RRF as
+measurement. A later channel or weight change recomputes evidence
+on the next GET; nothing is persisted. Event Lineage evidence
+remains the reconstruct/convex path.
+
+## References
+
+Cormack, G. V., Clarke, C. L. A., & Buettcher, S. (2009). Reciprocal
+rank fusion outperforms Condorcet and individual rank learning
+methods. In *Proceedings of the 32nd international ACM SIGIR
+conference on Research and development in information retrieval*
+(pp. 758–759). ACM. https://doi.org/10.1145/1571941.1572114
+
+Samuel, D., MacAvaney, S., Yates, A., Zhang, E., Zhang, S.,
+Macdonald, C., & Ounis, I. (2025). *Weighted reciprocal rank fusion
+for multi-channel retrieval* [Preprint].
+
+ADR 0024 (RankWeave fusion fail-closed)
+ADR 0124 (Event Lineage channel evidence; separate surface)

docs/adr/README.md

@@ -10,7 +10,7 @@ decision from them.
 | Supporting document | Normative ADR |
 |---|---|
 | [`product-technical-gap-baseline.md`](../product-technical-gap-baseline.md) | Product/technical traceability projection across the ADR set; ADRs remain normative |
-| [`lineage-bi-research-notes.md`](../lineage-bi-research-notes.md) | [0084](0084-lineage-research-grounding.md), [0062](0062-semantic-unit-embedding.md), [0064](0064-lineage-evidence-and-tree-assembly.md) |
+| [`lineage-bi-research-notes.md`](../lineage-bi-research-notes.md) | [0084](0084-lineage-research-grounding.md), [0062](0062-semantic-unit-embedding.md), [0064](0064-lineage-evidence-and-tree-assembly.md), [0024](0024-rankweave-fusion-fail-closed.md), [0167](0167-rankweave-ranking-channel-evidence.md) |
 | [`PROV_O_IMPLEMENTATION.md`](../PROV_O_IMPLEMENTATION.md) | [0065](0065-prov-o-provenance-boundary.md) |
 | [`PROV_O_IMPLEMENTATION_MATRIX.md`](../PROV_O_IMPLEMENTATION_MATRIX.md) | [0065](0065-prov-o-provenance-boundary.md) |
 | [`ONTOLOGY_NAMESPACE_INVENTORY.md`](../doctoring/ONTOLOGY_NAMESPACE_INVENTORY.md) | [0157](0157-public-ontology-namespace-identity.md) |

frontend/package.json

@@ -1,7 +1,7 @@
 {
   "name": "frontend",
   "private": true,
-  "version": "2.12.15",
+  "version": "2.12.17",
   "type": "module",
   "scripts": {
     "dev": "vite",

frontend/src/App.css

@@ -706,6 +706,27 @@
   gap: 0.5rem;
 }
 
+.ranking-hit {
+  flex-direction: column;
+  align-items: stretch;
+  justify-content: flex-start;
+}
+
+.ranking-channel-evidence-copy {
+  margin: 0 0 0.75rem;
+}
+
+.ranking-channel-evidence {
+  list-style: none;
+  margin: 0;
+  padding: 0 0 0 0.5rem;
+  font-size: 0.85rem;
+}
+
+.ranking-channel-evidence li {
+  margin: 0.15rem 0;
+}
+
 .ticket-title {
   flex: 1;
 }

frontend/src/App.test.tsx

@@ -78,6 +78,14 @@ describe("App, authenticated", () => {
         post_id: string;
         post_title: string;
         fused_rank: number;
+        channel_evidence?: {
+          signal_code: string;
+          signal_label: string;
+          channel_rank: number;
+          weight: number;
+          contribution: number;
+          rank: number;
+        }[];
       }[];
     };
     chatUnavailable?: boolean;
@@ -2592,11 +2600,47 @@ describe("App, authenticated", () => {
             post_id: "post-1",
             post_title: "Public post",
             fused_rank: 1,
+            channel_evidence: [
+              {
+                signal_code: "lexical",
+                signal_label: "Title overlap",
+                channel_rank: 2,
+                weight: 0.75,
+                contribution: 0.75 / 62,
+                rank: 1,
+              },
+              {
+                signal_code: "temporal",
+                signal_label: "Newest first",
+                channel_rank: 2,
+                weight: 0.25,
+                contribution: 0.25 / 62,
+                rank: 2,
+              },
+            ],
           },
           {
             post_id: "post-2",
             post_title: "Pricing renegotiation: revised quote sent",
             fused_rank: 2,
+            channel_evidence: [
+              {
+                signal_code: "lexical",
+                signal_label: "Title overlap",
+                channel_rank: 1,
+                weight: 0.75,
+                contribution: 0.75 / 61,
+                rank: 1,
+              },
+              {
+                signal_code: "temporal",
+                signal_label: "Newest first",
+                channel_rank: 1,
+                weight: 0.25,
+                contribution: 0.25 / 61,
+                rank: 2,
+              },
+            ],
           },
         ],
       },
@@ -2609,6 +2653,17 @@ describe("App, authenticated", () => {
     expect(rankingButton).toHaveTextContent("Public post");
     expect(rankingButton).toHaveTextContent("Rankings · rankweave");
     expect(rankingButton).toHaveTextContent("rank 1");
+    expect(
+      screen.getByText(
+        "RankWeave fused newest-first and title-overlap ranks. This is not a calibrated score.",
+      ),
+    ).toBeInTheDocument();
+    expect(
+      screen.getByRole("list", { name: "Ranking evidence for Public post" }),
+    ).toHaveTextContent("Title overlap rank 2, contribution 0.012097");
+    expect(
+      screen.getByRole("list", { name: "Ranking evidence for Public post" }),
+    ).toHaveTextContent("Newest first rank 2, contribution 0.004032");
     expect(screen.queryByRole("button", { name: /open ranking: private parent/i })).not.toBeInTheDocument();
 
     await userEvent.click(rankingButton);

frontend/src/App.tsx

@@ -3059,6 +3059,10 @@ function AnalysisRunsPanel({
   );
 }
 
+function formatRankingContribution(value: number): string {
+  return value.toFixed(6);
+}
+
 function RankingsPanel({
   accessToken,
   onSelectPost,
@@ -3077,9 +3081,9 @@ function RankingsPanel({
   }, [accessToken]);
 
   return (
-    <section className="popup-section lineage-home" aria-label="Rankings">
+    <section className="popup-section lineage-home" aria-label={t("Rankings")}>
       <div className="lineage-home-header">
-        <h2>Rankings</h2>
+        <h2>{t("Rankings")}</h2>
         {ranking && (
           <span className="post-badge">
             {ranking.status === "accepted"
@@ -3089,30 +3093,53 @@ function RankingsPanel({
         )}
       </div>
       {error && <p className="error">{error}</p>}
-      {ranking === null && !error && <p>Loading rankings...</p>}
+      {ranking === null && !error && <p>{t("Loading rankings...")}</p>}
       {ranking && ranking.status === "unavailable" && (
-        <p className="popup-placeholder">Rankings · RankWeave not available</p>
+        <p className="popup-placeholder">{t("Rankings · RankWeave not available")}</p>
       )}
       {ranking && ranking.status === "accepted" && ranking.rankings.length === 0 && (
-        <p className="popup-placeholder">No fused rankings from RankWeave.</p>
+        <p className="popup-placeholder">{t("No fused rankings from RankWeave.")}</p>
       )}
       {ranking && ranking.rankings.length > 0 && (
-        <ul className="ticket-list" aria-label="Fused rankings">
-          {ranking.rankings.map((hit) => (
-            <li key={hit.post_id} className="ticket-list-item">
-              <button
-                className="post-list-item"
-                aria-label={`Open ranking: ${hit.post_title}`}
-                onClick={() => onSelectPost(hit.post_id)}
-              >
-                <span className="ticket-title">{hit.post_title}</span>
-                <span className="post-badge">Rankings · rankweave</span>
-                <span className="post-badge">rank {hit.fused_rank}</span>
-              </button>
-            </li>
-          ))}
-        </ul>
-        )}
+        <>
+          <p className="ranking-channel-evidence-copy">
+            {t(
+              "RankWeave fused newest-first and title-overlap ranks. This is not a calibrated score.",
+            )}
+          </p>
+          <ul className="ticket-list" aria-label={t("Fused rankings")}>
+            {ranking.rankings.map((hit) => (
+              <li key={hit.post_id} className="ticket-list-item ranking-hit">
+                <button
+                  className="post-list-item"
+                  aria-label={tf("Open ranking: {title}", { title: hit.post_title })}
+                  onClick={() => onSelectPost(hit.post_id)}
+                >
+                  <span className="ticket-title">{hit.post_title}</span>
+                  <span className="post-badge">{t("Rankings · rankweave")}</span>
+                  <span className="post-badge">{tf("rank {rank}", { rank: String(hit.fused_rank) })}</span>
+                </button>
+                {(hit.channel_evidence ?? []).length > 0 ? (
+                  <ul
+                    className="ranking-channel-evidence"
+                    aria-label={tf("Ranking evidence for {title}", { title: hit.post_title })}
+                  >
+                    {(hit.channel_evidence ?? []).map((item) => (
+                      <li key={item.signal_code}>
+                        {tf("{label} rank {rank}, contribution {contribution}", {
+                          label: t(item.signal_label),
+                          rank: String(item.channel_rank),
+                          contribution: formatRankingContribution(item.contribution),
+                        })}
+                      </li>
+                    ))}
+                  </ul>
+                ) : null}
+              </li>
+            ))}
+          </ul>
+        </>
+      )}
     </section>
   );
 }

frontend/src/api.ts

@@ -1037,10 +1037,20 @@ export function startAnalysisRun(
   });
 }
 
+export interface RankingChannelEvidence {
+  signal_code: string;
+  signal_label: string;
+  channel_rank: number;
+  weight: number;
+  contribution: number;
+  rank: number;
+}
+
 export interface RankedPost {
   post_id: string;
   post_title: string;
   fused_rank: number;
+  channel_evidence?: RankingChannelEvidence[];
 }
 
 export interface RankingList {

frontend/src/i18n.test.ts

@@ -46,6 +46,9 @@ describe("i18n", () => {
     "Read leftover map rank {rank}, observed Y {observed}, and expected E {expected} after IRT main effects, then open this post.",
     "Leftover map rank 0 means no leftover structure after IRT main effects. Read observed Y {observed} and expected E {expected}, then open this post.",
     "Showing the first {shown} of {total} posts known at this cutoff.",
+    "Rankings",
+    "Title overlap",
+    "RankWeave fused newest-first and title-overlap ranks. This is not a calibrated score.",
   ] as const;
 
   it("supports the five product locales", () => {